xref: /netbsd-src/sys/nfs/nfs_subs.c (revision aa73cae19608873cc4d1f712c4a0f8f8435f1ffa)
1 /*	$NetBSD: nfs_subs.c,v 1.148 2005/02/26 22:39:50 perry Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)nfs_subs.c	8.8 (Berkeley) 5/22/95
35  */
36 
37 /*
38  * Copyright 2000 Wasabi Systems, Inc.
39  * All rights reserved.
40  *
41  * Written by Frank van der Linden for Wasabi Systems, Inc.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  * 3. All advertising materials mentioning features or use of this software
52  *    must display the following acknowledgement:
53  *      This product includes software developed for the NetBSD Project by
54  *      Wasabi Systems, Inc.
55  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
56  *    or promote products derived from this software without specific prior
57  *    written permission.
58  *
59  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
61  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
62  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
63  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
64  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
65  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
66  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
67  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
68  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
69  * POSSIBILITY OF SUCH DAMAGE.
70  */
71 
72 #include <sys/cdefs.h>
73 __KERNEL_RCSID(0, "$NetBSD: nfs_subs.c,v 1.148 2005/02/26 22:39:50 perry Exp $");
74 
75 #include "fs_nfs.h"
76 #include "opt_nfs.h"
77 #include "opt_nfsserver.h"
78 #include "opt_iso.h"
79 #include "opt_inet.h"
80 
81 /*
82  * These functions support the macros and help fiddle mbuf chains for
83  * the nfs op functions. They do things like create the rpc header and
84  * copy data between mbuf chains and uio lists.
85  */
86 #include <sys/param.h>
87 #include <sys/proc.h>
88 #include <sys/systm.h>
89 #include <sys/kernel.h>
90 #include <sys/mount.h>
91 #include <sys/vnode.h>
92 #include <sys/namei.h>
93 #include <sys/mbuf.h>
94 #include <sys/socket.h>
95 #include <sys/stat.h>
96 #include <sys/malloc.h>
97 #include <sys/filedesc.h>
98 #include <sys/time.h>
99 #include <sys/dirent.h>
100 
101 #include <uvm/uvm_extern.h>
102 
103 #include <nfs/rpcv2.h>
104 #include <nfs/nfsproto.h>
105 #include <nfs/nfsnode.h>
106 #include <nfs/nfs.h>
107 #include <nfs/xdr_subs.h>
108 #include <nfs/nfsm_subs.h>
109 #include <nfs/nfsmount.h>
110 #include <nfs/nqnfs.h>
111 #include <nfs/nfsrtt.h>
112 #include <nfs/nfs_var.h>
113 
114 #include <miscfs/specfs/specdev.h>
115 
116 #include <netinet/in.h>
117 #ifdef ISO
118 #include <netiso/iso.h>
119 #endif
120 
121 /*
122  * Data items converted to xdr at startup, since they are constant
123  * This is kinda hokey, but may save a little time doing byte swaps
124  */
125 u_int32_t nfs_xdrneg1;
126 u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr,
127 	rpc_mismatch, rpc_auth_unix, rpc_msgaccepted,
128 	rpc_auth_kerb;
129 u_int32_t nfs_prog, nqnfs_prog, nfs_true, nfs_false;
130 
131 /* And other global data */
132 const nfstype nfsv2_type[9] =
133 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON, NFCHR, NFNON };
134 const nfstype nfsv3_type[9] =
135 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK, NFFIFO, NFNON };
136 const enum vtype nv2tov_type[8] =
137 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON };
138 const enum vtype nv3tov_type[8] =
139 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO };
140 int nfs_ticks;
141 int nfs_commitsize;
142 
143 MALLOC_DEFINE(M_NFSDIROFF, "NFS diroff", "NFS directory cookies");
144 
145 /* NFS client/server stats. */
146 struct nfsstats nfsstats;
147 
148 /*
149  * Mapping of old NFS Version 2 RPC numbers to generic numbers.
150  */
151 const int nfsv3_procid[NFS_NPROCS] = {
152 	NFSPROC_NULL,
153 	NFSPROC_GETATTR,
154 	NFSPROC_SETATTR,
155 	NFSPROC_NOOP,
156 	NFSPROC_LOOKUP,
157 	NFSPROC_READLINK,
158 	NFSPROC_READ,
159 	NFSPROC_NOOP,
160 	NFSPROC_WRITE,
161 	NFSPROC_CREATE,
162 	NFSPROC_REMOVE,
163 	NFSPROC_RENAME,
164 	NFSPROC_LINK,
165 	NFSPROC_SYMLINK,
166 	NFSPROC_MKDIR,
167 	NFSPROC_RMDIR,
168 	NFSPROC_READDIR,
169 	NFSPROC_FSSTAT,
170 	NFSPROC_NOOP,
171 	NFSPROC_NOOP,
172 	NFSPROC_NOOP,
173 	NFSPROC_NOOP,
174 	NFSPROC_NOOP,
175 	NFSPROC_NOOP,
176 	NFSPROC_NOOP,
177 	NFSPROC_NOOP
178 };
179 
180 /*
181  * and the reverse mapping from generic to Version 2 procedure numbers
182  */
183 const int nfsv2_procid[NFS_NPROCS] = {
184 	NFSV2PROC_NULL,
185 	NFSV2PROC_GETATTR,
186 	NFSV2PROC_SETATTR,
187 	NFSV2PROC_LOOKUP,
188 	NFSV2PROC_NOOP,
189 	NFSV2PROC_READLINK,
190 	NFSV2PROC_READ,
191 	NFSV2PROC_WRITE,
192 	NFSV2PROC_CREATE,
193 	NFSV2PROC_MKDIR,
194 	NFSV2PROC_SYMLINK,
195 	NFSV2PROC_CREATE,
196 	NFSV2PROC_REMOVE,
197 	NFSV2PROC_RMDIR,
198 	NFSV2PROC_RENAME,
199 	NFSV2PROC_LINK,
200 	NFSV2PROC_READDIR,
201 	NFSV2PROC_NOOP,
202 	NFSV2PROC_STATFS,
203 	NFSV2PROC_NOOP,
204 	NFSV2PROC_NOOP,
205 	NFSV2PROC_NOOP,
206 	NFSV2PROC_NOOP,
207 	NFSV2PROC_NOOP,
208 	NFSV2PROC_NOOP,
209 	NFSV2PROC_NOOP,
210 };
211 
212 /*
213  * Maps errno values to nfs error numbers.
214  * Use NFSERR_IO as the catch all for ones not specifically defined in
215  * RFC 1094.
216  */
217 static const u_char nfsrv_v2errmap[ELAST] = {
218   NFSERR_PERM,	NFSERR_NOENT,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
219   NFSERR_NXIO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
220   NFSERR_IO,	NFSERR_IO,	NFSERR_ACCES,	NFSERR_IO,	NFSERR_IO,
221   NFSERR_IO,	NFSERR_EXIST,	NFSERR_IO,	NFSERR_NODEV,	NFSERR_NOTDIR,
222   NFSERR_ISDIR,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
223   NFSERR_IO,	NFSERR_FBIG,	NFSERR_NOSPC,	NFSERR_IO,	NFSERR_ROFS,
224   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
225   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
226   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
227   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
228   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
229   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
230   NFSERR_IO,	NFSERR_IO,	NFSERR_NAMETOL,	NFSERR_IO,	NFSERR_IO,
231   NFSERR_NOTEMPTY, NFSERR_IO,	NFSERR_IO,	NFSERR_DQUOT,	NFSERR_STALE,
232   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
233   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
234   NFSERR_IO,	NFSERR_IO,
235 };
236 
237 /*
238  * Maps errno values to nfs error numbers.
239  * Although it is not obvious whether or not NFS clients really care if
240  * a returned error value is in the specified list for the procedure, the
241  * safest thing to do is filter them appropriately. For Version 2, the
242  * X/Open XNFS document is the only specification that defines error values
243  * for each RPC (The RFC simply lists all possible error values for all RPCs),
244  * so I have decided to not do this for Version 2.
245  * The first entry is the default error return and the rest are the valid
246  * errors for that RPC in increasing numeric order.
247  */
248 static const short nfsv3err_null[] = {
249 	0,
250 	0,
251 };
252 
253 static const short nfsv3err_getattr[] = {
254 	NFSERR_IO,
255 	NFSERR_IO,
256 	NFSERR_STALE,
257 	NFSERR_BADHANDLE,
258 	NFSERR_SERVERFAULT,
259 	0,
260 };
261 
262 static const short nfsv3err_setattr[] = {
263 	NFSERR_IO,
264 	NFSERR_PERM,
265 	NFSERR_IO,
266 	NFSERR_ACCES,
267 	NFSERR_INVAL,
268 	NFSERR_NOSPC,
269 	NFSERR_ROFS,
270 	NFSERR_DQUOT,
271 	NFSERR_STALE,
272 	NFSERR_BADHANDLE,
273 	NFSERR_NOT_SYNC,
274 	NFSERR_SERVERFAULT,
275 	0,
276 };
277 
278 static const short nfsv3err_lookup[] = {
279 	NFSERR_IO,
280 	NFSERR_NOENT,
281 	NFSERR_IO,
282 	NFSERR_ACCES,
283 	NFSERR_NOTDIR,
284 	NFSERR_NAMETOL,
285 	NFSERR_STALE,
286 	NFSERR_BADHANDLE,
287 	NFSERR_SERVERFAULT,
288 	0,
289 };
290 
291 static const short nfsv3err_access[] = {
292 	NFSERR_IO,
293 	NFSERR_IO,
294 	NFSERR_STALE,
295 	NFSERR_BADHANDLE,
296 	NFSERR_SERVERFAULT,
297 	0,
298 };
299 
300 static const short nfsv3err_readlink[] = {
301 	NFSERR_IO,
302 	NFSERR_IO,
303 	NFSERR_ACCES,
304 	NFSERR_INVAL,
305 	NFSERR_STALE,
306 	NFSERR_BADHANDLE,
307 	NFSERR_NOTSUPP,
308 	NFSERR_SERVERFAULT,
309 	0,
310 };
311 
312 static const short nfsv3err_read[] = {
313 	NFSERR_IO,
314 	NFSERR_IO,
315 	NFSERR_NXIO,
316 	NFSERR_ACCES,
317 	NFSERR_INVAL,
318 	NFSERR_STALE,
319 	NFSERR_BADHANDLE,
320 	NFSERR_SERVERFAULT,
321 	NFSERR_JUKEBOX,
322 	0,
323 };
324 
325 static const short nfsv3err_write[] = {
326 	NFSERR_IO,
327 	NFSERR_IO,
328 	NFSERR_ACCES,
329 	NFSERR_INVAL,
330 	NFSERR_FBIG,
331 	NFSERR_NOSPC,
332 	NFSERR_ROFS,
333 	NFSERR_DQUOT,
334 	NFSERR_STALE,
335 	NFSERR_BADHANDLE,
336 	NFSERR_SERVERFAULT,
337 	NFSERR_JUKEBOX,
338 	0,
339 };
340 
341 static const short nfsv3err_create[] = {
342 	NFSERR_IO,
343 	NFSERR_IO,
344 	NFSERR_ACCES,
345 	NFSERR_EXIST,
346 	NFSERR_NOTDIR,
347 	NFSERR_NOSPC,
348 	NFSERR_ROFS,
349 	NFSERR_NAMETOL,
350 	NFSERR_DQUOT,
351 	NFSERR_STALE,
352 	NFSERR_BADHANDLE,
353 	NFSERR_NOTSUPP,
354 	NFSERR_SERVERFAULT,
355 	0,
356 };
357 
358 static const short nfsv3err_mkdir[] = {
359 	NFSERR_IO,
360 	NFSERR_IO,
361 	NFSERR_ACCES,
362 	NFSERR_EXIST,
363 	NFSERR_NOTDIR,
364 	NFSERR_NOSPC,
365 	NFSERR_ROFS,
366 	NFSERR_NAMETOL,
367 	NFSERR_DQUOT,
368 	NFSERR_STALE,
369 	NFSERR_BADHANDLE,
370 	NFSERR_NOTSUPP,
371 	NFSERR_SERVERFAULT,
372 	0,
373 };
374 
375 static const short nfsv3err_symlink[] = {
376 	NFSERR_IO,
377 	NFSERR_IO,
378 	NFSERR_ACCES,
379 	NFSERR_EXIST,
380 	NFSERR_NOTDIR,
381 	NFSERR_NOSPC,
382 	NFSERR_ROFS,
383 	NFSERR_NAMETOL,
384 	NFSERR_DQUOT,
385 	NFSERR_STALE,
386 	NFSERR_BADHANDLE,
387 	NFSERR_NOTSUPP,
388 	NFSERR_SERVERFAULT,
389 	0,
390 };
391 
392 static const short nfsv3err_mknod[] = {
393 	NFSERR_IO,
394 	NFSERR_IO,
395 	NFSERR_ACCES,
396 	NFSERR_EXIST,
397 	NFSERR_NOTDIR,
398 	NFSERR_NOSPC,
399 	NFSERR_ROFS,
400 	NFSERR_NAMETOL,
401 	NFSERR_DQUOT,
402 	NFSERR_STALE,
403 	NFSERR_BADHANDLE,
404 	NFSERR_NOTSUPP,
405 	NFSERR_SERVERFAULT,
406 	NFSERR_BADTYPE,
407 	0,
408 };
409 
410 static const short nfsv3err_remove[] = {
411 	NFSERR_IO,
412 	NFSERR_NOENT,
413 	NFSERR_IO,
414 	NFSERR_ACCES,
415 	NFSERR_NOTDIR,
416 	NFSERR_ROFS,
417 	NFSERR_NAMETOL,
418 	NFSERR_STALE,
419 	NFSERR_BADHANDLE,
420 	NFSERR_SERVERFAULT,
421 	0,
422 };
423 
424 static const short nfsv3err_rmdir[] = {
425 	NFSERR_IO,
426 	NFSERR_NOENT,
427 	NFSERR_IO,
428 	NFSERR_ACCES,
429 	NFSERR_EXIST,
430 	NFSERR_NOTDIR,
431 	NFSERR_INVAL,
432 	NFSERR_ROFS,
433 	NFSERR_NAMETOL,
434 	NFSERR_NOTEMPTY,
435 	NFSERR_STALE,
436 	NFSERR_BADHANDLE,
437 	NFSERR_NOTSUPP,
438 	NFSERR_SERVERFAULT,
439 	0,
440 };
441 
442 static const short nfsv3err_rename[] = {
443 	NFSERR_IO,
444 	NFSERR_NOENT,
445 	NFSERR_IO,
446 	NFSERR_ACCES,
447 	NFSERR_EXIST,
448 	NFSERR_XDEV,
449 	NFSERR_NOTDIR,
450 	NFSERR_ISDIR,
451 	NFSERR_INVAL,
452 	NFSERR_NOSPC,
453 	NFSERR_ROFS,
454 	NFSERR_MLINK,
455 	NFSERR_NAMETOL,
456 	NFSERR_NOTEMPTY,
457 	NFSERR_DQUOT,
458 	NFSERR_STALE,
459 	NFSERR_BADHANDLE,
460 	NFSERR_NOTSUPP,
461 	NFSERR_SERVERFAULT,
462 	0,
463 };
464 
465 static const short nfsv3err_link[] = {
466 	NFSERR_IO,
467 	NFSERR_IO,
468 	NFSERR_ACCES,
469 	NFSERR_EXIST,
470 	NFSERR_XDEV,
471 	NFSERR_NOTDIR,
472 	NFSERR_INVAL,
473 	NFSERR_NOSPC,
474 	NFSERR_ROFS,
475 	NFSERR_MLINK,
476 	NFSERR_NAMETOL,
477 	NFSERR_DQUOT,
478 	NFSERR_STALE,
479 	NFSERR_BADHANDLE,
480 	NFSERR_NOTSUPP,
481 	NFSERR_SERVERFAULT,
482 	0,
483 };
484 
485 static const short nfsv3err_readdir[] = {
486 	NFSERR_IO,
487 	NFSERR_IO,
488 	NFSERR_ACCES,
489 	NFSERR_NOTDIR,
490 	NFSERR_STALE,
491 	NFSERR_BADHANDLE,
492 	NFSERR_BAD_COOKIE,
493 	NFSERR_TOOSMALL,
494 	NFSERR_SERVERFAULT,
495 	0,
496 };
497 
498 static const short nfsv3err_readdirplus[] = {
499 	NFSERR_IO,
500 	NFSERR_IO,
501 	NFSERR_ACCES,
502 	NFSERR_NOTDIR,
503 	NFSERR_STALE,
504 	NFSERR_BADHANDLE,
505 	NFSERR_BAD_COOKIE,
506 	NFSERR_NOTSUPP,
507 	NFSERR_TOOSMALL,
508 	NFSERR_SERVERFAULT,
509 	0,
510 };
511 
512 static const short nfsv3err_fsstat[] = {
513 	NFSERR_IO,
514 	NFSERR_IO,
515 	NFSERR_STALE,
516 	NFSERR_BADHANDLE,
517 	NFSERR_SERVERFAULT,
518 	0,
519 };
520 
521 static const short nfsv3err_fsinfo[] = {
522 	NFSERR_STALE,
523 	NFSERR_STALE,
524 	NFSERR_BADHANDLE,
525 	NFSERR_SERVERFAULT,
526 	0,
527 };
528 
529 static const short nfsv3err_pathconf[] = {
530 	NFSERR_STALE,
531 	NFSERR_STALE,
532 	NFSERR_BADHANDLE,
533 	NFSERR_SERVERFAULT,
534 	0,
535 };
536 
537 static const short nfsv3err_commit[] = {
538 	NFSERR_IO,
539 	NFSERR_IO,
540 	NFSERR_STALE,
541 	NFSERR_BADHANDLE,
542 	NFSERR_SERVERFAULT,
543 	0,
544 };
545 
546 static const short * const nfsrv_v3errmap[] = {
547 	nfsv3err_null,
548 	nfsv3err_getattr,
549 	nfsv3err_setattr,
550 	nfsv3err_lookup,
551 	nfsv3err_access,
552 	nfsv3err_readlink,
553 	nfsv3err_read,
554 	nfsv3err_write,
555 	nfsv3err_create,
556 	nfsv3err_mkdir,
557 	nfsv3err_symlink,
558 	nfsv3err_mknod,
559 	nfsv3err_remove,
560 	nfsv3err_rmdir,
561 	nfsv3err_rename,
562 	nfsv3err_link,
563 	nfsv3err_readdir,
564 	nfsv3err_readdirplus,
565 	nfsv3err_fsstat,
566 	nfsv3err_fsinfo,
567 	nfsv3err_pathconf,
568 	nfsv3err_commit,
569 };
570 
571 extern struct nfsrtt nfsrtt;
572 extern time_t nqnfsstarttime;
573 extern int nqsrv_clockskew;
574 extern int nqsrv_writeslack;
575 extern int nqsrv_maxlease;
576 extern const int nqnfs_piggy[NFS_NPROCS];
577 extern struct nfsnodehashhead *nfsnodehashtbl;
578 extern u_long nfsnodehash;
579 
580 u_long nfsdirhashmask;
581 
582 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *));
583 
584 /*
585  * Create the header for an rpc request packet
586  * The hsiz is the size of the rest of the nfs request header.
587  * (just used to decide if a cluster is a good idea)
588  */
589 struct mbuf *
590 nfsm_reqh(np, procid, hsiz, bposp)
591 	struct nfsnode *np;
592 	u_long procid;
593 	int hsiz;
594 	caddr_t *bposp;
595 {
596 	struct mbuf *mb;
597 	caddr_t bpos;
598 #ifndef NFS_V2_ONLY
599 	struct nfsmount *nmp;
600 	u_int32_t *tl;
601 	int nqflag;
602 #endif
603 
604 	mb = m_get(M_WAIT, MT_DATA);
605 	MCLAIM(mb, &nfs_mowner);
606 	if (hsiz >= MINCLSIZE)
607 		m_clget(mb, M_WAIT);
608 	mb->m_len = 0;
609 	bpos = mtod(mb, caddr_t);
610 
611 #ifndef NFS_V2_ONLY
612 	/*
613 	 * For NQNFS, add lease request.
614 	 */
615 	if (np) {
616 		nmp = VFSTONFS(np->n_vnode->v_mount);
617 		if (nmp->nm_flag & NFSMNT_NQNFS) {
618 			nqflag = NQNFS_NEEDLEASE(np, procid);
619 			if (nqflag) {
620 				nfsm_build(tl, u_int32_t *, 2*NFSX_UNSIGNED);
621 				*tl++ = txdr_unsigned(nqflag);
622 				*tl = txdr_unsigned(nmp->nm_leaseterm);
623 			} else {
624 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
625 				*tl = 0;
626 			}
627 		}
628 	}
629 #endif
630 	/* Finally, return values */
631 	*bposp = bpos;
632 	return (mb);
633 }
634 
635 /*
636  * Build the RPC header and fill in the authorization info.
637  * The authorization string argument is only used when the credentials
638  * come from outside of the kernel.
639  * Returns the head of the mbuf list.
640  */
641 struct mbuf *
642 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len,
643 	verf_str, mrest, mrest_len, mbp, xidp)
644 	struct ucred *cr;
645 	int nmflag;
646 	int procid;
647 	int auth_type;
648 	int auth_len;
649 	char *auth_str;
650 	int verf_len;
651 	char *verf_str;
652 	struct mbuf *mrest;
653 	int mrest_len;
654 	struct mbuf **mbp;
655 	u_int32_t *xidp;
656 {
657 	struct mbuf *mb;
658 	u_int32_t *tl;
659 	caddr_t bpos;
660 	int i;
661 	struct mbuf *mreq;
662 	int siz, grpsiz, authsiz;
663 
664 	authsiz = nfsm_rndup(auth_len);
665 	mb = m_gethdr(M_WAIT, MT_DATA);
666 	MCLAIM(mb, &nfs_mowner);
667 	if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) {
668 		m_clget(mb, M_WAIT);
669 	} else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) {
670 		MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED);
671 	} else {
672 		MH_ALIGN(mb, 8 * NFSX_UNSIGNED);
673 	}
674 	mb->m_len = 0;
675 	mreq = mb;
676 	bpos = mtod(mb, caddr_t);
677 
678 	/*
679 	 * First the RPC header.
680 	 */
681 	nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
682 
683 	*tl++ = *xidp = nfs_getxid();
684 	*tl++ = rpc_call;
685 	*tl++ = rpc_vers;
686 	if (nmflag & NFSMNT_NQNFS) {
687 		*tl++ = txdr_unsigned(NQNFS_PROG);
688 		*tl++ = txdr_unsigned(NQNFS_VER3);
689 	} else {
690 		*tl++ = txdr_unsigned(NFS_PROG);
691 		if (nmflag & NFSMNT_NFSV3)
692 			*tl++ = txdr_unsigned(NFS_VER3);
693 		else
694 			*tl++ = txdr_unsigned(NFS_VER2);
695 	}
696 	if (nmflag & NFSMNT_NFSV3)
697 		*tl++ = txdr_unsigned(procid);
698 	else
699 		*tl++ = txdr_unsigned(nfsv2_procid[procid]);
700 
701 	/*
702 	 * And then the authorization cred.
703 	 */
704 	*tl++ = txdr_unsigned(auth_type);
705 	*tl = txdr_unsigned(authsiz);
706 	switch (auth_type) {
707 	case RPCAUTH_UNIX:
708 		nfsm_build(tl, u_int32_t *, auth_len);
709 		*tl++ = 0;		/* stamp ?? */
710 		*tl++ = 0;		/* NULL hostname */
711 		*tl++ = txdr_unsigned(cr->cr_uid);
712 		*tl++ = txdr_unsigned(cr->cr_gid);
713 		grpsiz = (auth_len >> 2) - 5;
714 		*tl++ = txdr_unsigned(grpsiz);
715 		for (i = 0; i < grpsiz; i++)
716 			*tl++ = txdr_unsigned(cr->cr_groups[i]);
717 		break;
718 	case RPCAUTH_KERB4:
719 		siz = auth_len;
720 		while (siz > 0) {
721 			if (M_TRAILINGSPACE(mb) == 0) {
722 				struct mbuf *mb2;
723 				mb2 = m_get(M_WAIT, MT_DATA);
724 				MCLAIM(mb2, &nfs_mowner);
725 				if (siz >= MINCLSIZE)
726 					m_clget(mb2, M_WAIT);
727 				mb->m_next = mb2;
728 				mb = mb2;
729 				mb->m_len = 0;
730 				bpos = mtod(mb, caddr_t);
731 			}
732 			i = min(siz, M_TRAILINGSPACE(mb));
733 			memcpy(bpos, auth_str, i);
734 			mb->m_len += i;
735 			auth_str += i;
736 			bpos += i;
737 			siz -= i;
738 		}
739 		if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) {
740 			for (i = 0; i < siz; i++)
741 				*bpos++ = '\0';
742 			mb->m_len += siz;
743 		}
744 		break;
745 	};
746 
747 	/*
748 	 * And the verifier...
749 	 */
750 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
751 	if (verf_str) {
752 		*tl++ = txdr_unsigned(RPCAUTH_KERB4);
753 		*tl = txdr_unsigned(verf_len);
754 		siz = verf_len;
755 		while (siz > 0) {
756 			if (M_TRAILINGSPACE(mb) == 0) {
757 				struct mbuf *mb2;
758 				mb2 = m_get(M_WAIT, MT_DATA);
759 				MCLAIM(mb2, &nfs_mowner);
760 				if (siz >= MINCLSIZE)
761 					m_clget(mb2, M_WAIT);
762 				mb->m_next = mb2;
763 				mb = mb2;
764 				mb->m_len = 0;
765 				bpos = mtod(mb, caddr_t);
766 			}
767 			i = min(siz, M_TRAILINGSPACE(mb));
768 			memcpy(bpos, verf_str, i);
769 			mb->m_len += i;
770 			verf_str += i;
771 			bpos += i;
772 			siz -= i;
773 		}
774 		if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) {
775 			for (i = 0; i < siz; i++)
776 				*bpos++ = '\0';
777 			mb->m_len += siz;
778 		}
779 	} else {
780 		*tl++ = txdr_unsigned(RPCAUTH_NULL);
781 		*tl = 0;
782 	}
783 	mb->m_next = mrest;
784 	mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len;
785 	mreq->m_pkthdr.rcvif = (struct ifnet *)0;
786 	*mbp = mb;
787 	return (mreq);
788 }
789 
790 /*
791  * copies mbuf chain to the uio scatter/gather list
792  */
793 int
794 nfsm_mbuftouio(mrep, uiop, siz, dpos)
795 	struct mbuf **mrep;
796 	struct uio *uiop;
797 	int siz;
798 	caddr_t *dpos;
799 {
800 	char *mbufcp, *uiocp;
801 	int xfer, left, len;
802 	struct mbuf *mp;
803 	long uiosiz, rem;
804 	int error = 0;
805 
806 	mp = *mrep;
807 	mbufcp = *dpos;
808 	len = mtod(mp, caddr_t)+mp->m_len-mbufcp;
809 	rem = nfsm_rndup(siz)-siz;
810 	while (siz > 0) {
811 		if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL)
812 			return (EFBIG);
813 		left = uiop->uio_iov->iov_len;
814 		uiocp = uiop->uio_iov->iov_base;
815 		if (left > siz)
816 			left = siz;
817 		uiosiz = left;
818 		while (left > 0) {
819 			while (len == 0) {
820 				mp = mp->m_next;
821 				if (mp == NULL)
822 					return (EBADRPC);
823 				mbufcp = mtod(mp, caddr_t);
824 				len = mp->m_len;
825 			}
826 			xfer = (left > len) ? len : left;
827 #ifdef notdef
828 			/* Not Yet.. */
829 			if (uiop->uio_iov->iov_op != NULL)
830 				(*(uiop->uio_iov->iov_op))
831 				(mbufcp, uiocp, xfer);
832 			else
833 #endif
834 			if (uiop->uio_segflg == UIO_SYSSPACE)
835 				memcpy(uiocp, mbufcp, xfer);
836 			else
837 				copyout(mbufcp, uiocp, xfer);
838 			left -= xfer;
839 			len -= xfer;
840 			mbufcp += xfer;
841 			uiocp += xfer;
842 			uiop->uio_offset += xfer;
843 			uiop->uio_resid -= xfer;
844 		}
845 		if (uiop->uio_iov->iov_len <= siz) {
846 			uiop->uio_iovcnt--;
847 			uiop->uio_iov++;
848 		} else {
849 			uiop->uio_iov->iov_base =
850 			    (caddr_t)uiop->uio_iov->iov_base + uiosiz;
851 			uiop->uio_iov->iov_len -= uiosiz;
852 		}
853 		siz -= uiosiz;
854 	}
855 	*dpos = mbufcp;
856 	*mrep = mp;
857 	if (rem > 0) {
858 		if (len < rem)
859 			error = nfs_adv(mrep, dpos, rem, len);
860 		else
861 			*dpos += rem;
862 	}
863 	return (error);
864 }
865 
866 /*
867  * copies a uio scatter/gather list to an mbuf chain.
868  * NOTE: can ony handle iovcnt == 1
869  */
870 int
871 nfsm_uiotombuf(uiop, mq, siz, bpos)
872 	struct uio *uiop;
873 	struct mbuf **mq;
874 	int siz;
875 	caddr_t *bpos;
876 {
877 	char *uiocp;
878 	struct mbuf *mp, *mp2;
879 	int xfer, left, mlen;
880 	int uiosiz, clflg, rem;
881 	char *cp;
882 
883 #ifdef DIAGNOSTIC
884 	if (uiop->uio_iovcnt != 1)
885 		panic("nfsm_uiotombuf: iovcnt != 1");
886 #endif
887 
888 	if (siz > MLEN)		/* or should it >= MCLBYTES ?? */
889 		clflg = 1;
890 	else
891 		clflg = 0;
892 	rem = nfsm_rndup(siz)-siz;
893 	mp = mp2 = *mq;
894 	while (siz > 0) {
895 		left = uiop->uio_iov->iov_len;
896 		uiocp = uiop->uio_iov->iov_base;
897 		if (left > siz)
898 			left = siz;
899 		uiosiz = left;
900 		while (left > 0) {
901 			mlen = M_TRAILINGSPACE(mp);
902 			if (mlen == 0) {
903 				mp = m_get(M_WAIT, MT_DATA);
904 				MCLAIM(mp, &nfs_mowner);
905 				if (clflg)
906 					m_clget(mp, M_WAIT);
907 				mp->m_len = 0;
908 				mp2->m_next = mp;
909 				mp2 = mp;
910 				mlen = M_TRAILINGSPACE(mp);
911 			}
912 			xfer = (left > mlen) ? mlen : left;
913 #ifdef notdef
914 			/* Not Yet.. */
915 			if (uiop->uio_iov->iov_op != NULL)
916 				(*(uiop->uio_iov->iov_op))
917 				(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
918 			else
919 #endif
920 			if (uiop->uio_segflg == UIO_SYSSPACE)
921 				memcpy(mtod(mp, caddr_t)+mp->m_len, uiocp, xfer);
922 			else
923 				copyin(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
924 			mp->m_len += xfer;
925 			left -= xfer;
926 			uiocp += xfer;
927 			uiop->uio_offset += xfer;
928 			uiop->uio_resid -= xfer;
929 		}
930 		uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
931 		    uiosiz;
932 		uiop->uio_iov->iov_len -= uiosiz;
933 		siz -= uiosiz;
934 	}
935 	if (rem > 0) {
936 		if (rem > M_TRAILINGSPACE(mp)) {
937 			mp = m_get(M_WAIT, MT_DATA);
938 			MCLAIM(mp, &nfs_mowner);
939 			mp->m_len = 0;
940 			mp2->m_next = mp;
941 		}
942 		cp = mtod(mp, caddr_t)+mp->m_len;
943 		for (left = 0; left < rem; left++)
944 			*cp++ = '\0';
945 		mp->m_len += rem;
946 		*bpos = cp;
947 	} else
948 		*bpos = mtod(mp, caddr_t)+mp->m_len;
949 	*mq = mp;
950 	return (0);
951 }
952 
953 /*
954  * Get at least "siz" bytes of correctly aligned data.
955  * When called the mbuf pointers are not necessarily correct,
956  * dsosp points to what ought to be in m_data and left contains
957  * what ought to be in m_len.
958  * This is used by the macros nfsm_dissect and nfsm_dissecton for tough
959  * cases. (The macros use the vars. dpos and dpos2)
960  */
961 int
962 nfsm_disct(mdp, dposp, siz, left, cp2)
963 	struct mbuf **mdp;
964 	caddr_t *dposp;
965 	int siz;
966 	int left;
967 	caddr_t *cp2;
968 {
969 	struct mbuf *m1, *m2;
970 	struct mbuf *havebuf = NULL;
971 	caddr_t src = *dposp;
972 	caddr_t dst;
973 	int len;
974 
975 #ifdef DEBUG
976 	if (left < 0)
977 		panic("nfsm_disct: left < 0");
978 #endif
979 	m1 = *mdp;
980 	/*
981 	 * Skip through the mbuf chain looking for an mbuf with
982 	 * some data. If the first mbuf found has enough data
983 	 * and it is correctly aligned return it.
984 	 */
985 	while (left == 0) {
986 		havebuf = m1;
987 		*mdp = m1 = m1->m_next;
988 		if (m1 == NULL)
989 			return (EBADRPC);
990 		src = mtod(m1, caddr_t);
991 		left = m1->m_len;
992 		/*
993 		 * If we start a new mbuf and it is big enough
994 		 * and correctly aligned just return it, don't
995 		 * do any pull up.
996 		 */
997 		if (left >= siz && nfsm_aligned(src)) {
998 			*cp2 = src;
999 			*dposp = src + siz;
1000 			return (0);
1001 		}
1002 	}
1003 	if (m1->m_flags & M_EXT) {
1004 		if (havebuf) {
1005 			/* If the first mbuf with data has external data
1006 			 * and there is a previous empty mbuf use it
1007 			 * to move the data into.
1008 			 */
1009 			m2 = m1;
1010 			*mdp = m1 = havebuf;
1011 			if (m1->m_flags & M_EXT) {
1012 				MEXTREMOVE(m1);
1013 			}
1014 		} else {
1015 			/*
1016 			 * If the first mbuf has a external data
1017 			 * and there is no previous empty mbuf
1018 			 * allocate a new mbuf and move the external
1019 			 * data to the new mbuf. Also make the first
1020 			 * mbuf look empty.
1021 			 */
1022 			m2 = m_get(M_WAIT, MT_DATA);
1023 			m2->m_ext = m1->m_ext;
1024 			m2->m_data = src;
1025 			m2->m_len = left;
1026 			MCLADDREFERENCE(m1, m2);
1027 			MEXTREMOVE(m1);
1028 			m2->m_next = m1->m_next;
1029 			m1->m_next = m2;
1030 		}
1031 		m1->m_len = 0;
1032 		if (m1->m_flags & M_PKTHDR)
1033 			dst = m1->m_pktdat;
1034 		else
1035 			dst = m1->m_dat;
1036 		m1->m_data = dst;
1037 	} else {
1038 		/*
1039 		 * If the first mbuf has no external data
1040 		 * move the data to the front of the mbuf.
1041 		 */
1042 		if (m1->m_flags & M_PKTHDR)
1043 			dst = m1->m_pktdat;
1044 		else
1045 			dst = m1->m_dat;
1046 		m1->m_data = dst;
1047 		if (dst != src)
1048 			memmove(dst, src, left);
1049 		dst += left;
1050 		m1->m_len = left;
1051 		m2 = m1->m_next;
1052 	}
1053 	*cp2 = m1->m_data;
1054 	*dposp = mtod(m1, caddr_t) + siz;
1055 	/*
1056 	 * Loop through mbufs pulling data up into first mbuf until
1057 	 * the first mbuf is full or there is no more data to
1058 	 * pullup.
1059 	 */
1060 	while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) {
1061 		if ((len = min(len, m2->m_len)) != 0)
1062 			memcpy(dst, m2->m_data, len);
1063 		m1->m_len += len;
1064 		dst += len;
1065 		m2->m_data += len;
1066 		m2->m_len -= len;
1067 		m2 = m2->m_next;
1068 	}
1069 	if (m1->m_len < siz)
1070 		return (EBADRPC);
1071 	return (0);
1072 }
1073 
1074 /*
1075  * Advance the position in the mbuf chain.
1076  */
1077 int
1078 nfs_adv(mdp, dposp, offs, left)
1079 	struct mbuf **mdp;
1080 	caddr_t *dposp;
1081 	int offs;
1082 	int left;
1083 {
1084 	struct mbuf *m;
1085 	int s;
1086 
1087 	m = *mdp;
1088 	s = left;
1089 	while (s < offs) {
1090 		offs -= s;
1091 		m = m->m_next;
1092 		if (m == NULL)
1093 			return (EBADRPC);
1094 		s = m->m_len;
1095 	}
1096 	*mdp = m;
1097 	*dposp = mtod(m, caddr_t)+offs;
1098 	return (0);
1099 }
1100 
1101 /*
1102  * Copy a string into mbufs for the hard cases...
1103  */
1104 int
1105 nfsm_strtmbuf(mb, bpos, cp, siz)
1106 	struct mbuf **mb;
1107 	char **bpos;
1108 	const char *cp;
1109 	long siz;
1110 {
1111 	struct mbuf *m1 = NULL, *m2;
1112 	long left, xfer, len, tlen;
1113 	u_int32_t *tl;
1114 	int putsize;
1115 
1116 	putsize = 1;
1117 	m2 = *mb;
1118 	left = M_TRAILINGSPACE(m2);
1119 	if (left > 0) {
1120 		tl = ((u_int32_t *)(*bpos));
1121 		*tl++ = txdr_unsigned(siz);
1122 		putsize = 0;
1123 		left -= NFSX_UNSIGNED;
1124 		m2->m_len += NFSX_UNSIGNED;
1125 		if (left > 0) {
1126 			memcpy((caddr_t) tl, cp, left);
1127 			siz -= left;
1128 			cp += left;
1129 			m2->m_len += left;
1130 			left = 0;
1131 		}
1132 	}
1133 	/* Loop around adding mbufs */
1134 	while (siz > 0) {
1135 		m1 = m_get(M_WAIT, MT_DATA);
1136 		MCLAIM(m1, &nfs_mowner);
1137 		if (siz > MLEN)
1138 			m_clget(m1, M_WAIT);
1139 		m1->m_len = NFSMSIZ(m1);
1140 		m2->m_next = m1;
1141 		m2 = m1;
1142 		tl = mtod(m1, u_int32_t *);
1143 		tlen = 0;
1144 		if (putsize) {
1145 			*tl++ = txdr_unsigned(siz);
1146 			m1->m_len -= NFSX_UNSIGNED;
1147 			tlen = NFSX_UNSIGNED;
1148 			putsize = 0;
1149 		}
1150 		if (siz < m1->m_len) {
1151 			len = nfsm_rndup(siz);
1152 			xfer = siz;
1153 			if (xfer < len)
1154 				*(tl+(xfer>>2)) = 0;
1155 		} else {
1156 			xfer = len = m1->m_len;
1157 		}
1158 		memcpy((caddr_t) tl, cp, xfer);
1159 		m1->m_len = len+tlen;
1160 		siz -= xfer;
1161 		cp += xfer;
1162 	}
1163 	*mb = m1;
1164 	*bpos = mtod(m1, caddr_t)+m1->m_len;
1165 	return (0);
1166 }
1167 
1168 /*
1169  * Directory caching routines. They work as follows:
1170  * - a cache is maintained per VDIR nfsnode.
1171  * - for each offset cookie that is exported to userspace, and can
1172  *   thus be thrown back at us as an offset to VOP_READDIR, store
1173  *   information in the cache.
1174  * - cached are:
1175  *   - cookie itself
1176  *   - blocknumber (essentially just a search key in the buffer cache)
1177  *   - entry number in block.
1178  *   - offset cookie of block in which this entry is stored
1179  *   - 32 bit cookie if NFSMNT_XLATECOOKIE is used.
1180  * - entries are looked up in a hash table
1181  * - also maintained is an LRU list of entries, used to determine
1182  *   which ones to delete if the cache grows too large.
1183  * - if 32 <-> 64 translation mode is requested for a filesystem,
1184  *   the cache also functions as a translation table
1185  * - in the translation case, invalidating the cache does not mean
1186  *   flushing it, but just marking entries as invalid, except for
1187  *   the <64bit cookie, 32bitcookie> pair which is still valid, to
1188  *   still be able to use the cache as a translation table.
1189  * - 32 bit cookies are uniquely created by combining the hash table
1190  *   entry value, and one generation count per hash table entry,
1191  *   incremented each time an entry is appended to the chain.
1192  * - the cache is invalidated each time a direcory is modified
1193  * - sanity checks are also done; if an entry in a block turns
1194  *   out not to have a matching cookie, the cache is invalidated
1195  *   and a new block starting from the wanted offset is fetched from
1196  *   the server.
1197  * - directory entries as read from the server are extended to contain
1198  *   the 64bit and, optionally, the 32bit cookies, for sanity checking
1199  *   the cache and exporting them to userspace through the cookie
1200  *   argument to VOP_READDIR.
1201  */
1202 
1203 u_long
1204 nfs_dirhash(off)
1205 	off_t off;
1206 {
1207 	int i;
1208 	char *cp = (char *)&off;
1209 	u_long sum = 0L;
1210 
1211 	for (i = 0 ; i < sizeof (off); i++)
1212 		sum += *cp++;
1213 
1214 	return sum;
1215 }
1216 
1217 #define	_NFSDC_MTX(np)		(&NFSTOV(np)->v_interlock)
1218 #define	NFSDC_LOCK(np)		simple_lock(_NFSDC_MTX(np))
1219 #define	NFSDC_UNLOCK(np)	simple_unlock(_NFSDC_MTX(np))
1220 #define	NFSDC_ASSERT_LOCKED(np) LOCK_ASSERT(simple_lock_held(_NFSDC_MTX(np)))
1221 
1222 void
1223 nfs_initdircache(vp)
1224 	struct vnode *vp;
1225 {
1226 	struct nfsnode *np = VTONFS(vp);
1227 	struct nfsdirhashhead *dircache;
1228 
1229 	dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, M_NFSDIROFF,
1230 	    M_WAITOK, &nfsdirhashmask);
1231 
1232 	NFSDC_LOCK(np);
1233 	if (np->n_dircache == NULL) {
1234 		np->n_dircachesize = 0;
1235 		np->n_dircache = dircache;
1236 		dircache = NULL;
1237 		TAILQ_INIT(&np->n_dirchain);
1238 	}
1239 	NFSDC_UNLOCK(np);
1240 	if (dircache)
1241 		hashdone(dircache, M_NFSDIROFF);
1242 }
1243 
1244 void
1245 nfs_initdirxlatecookie(vp)
1246 	struct vnode *vp;
1247 {
1248 	struct nfsnode *np = VTONFS(vp);
1249 	unsigned *dirgens;
1250 
1251 	KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE);
1252 
1253 	dirgens = malloc(NFS_DIRHASHSIZ * sizeof (unsigned), M_NFSDIROFF,
1254 	    M_WAITOK|M_ZERO);
1255 	NFSDC_LOCK(np);
1256 	if (np->n_dirgens == NULL) {
1257 		np->n_dirgens = dirgens;
1258 		dirgens = NULL;
1259 	}
1260 	NFSDC_UNLOCK(np);
1261 	if (dirgens)
1262 		free(dirgens, M_NFSDIROFF);
1263 }
1264 
1265 static const struct nfsdircache dzero;
1266 
1267 static void nfs_unlinkdircache __P((struct nfsnode *np, struct nfsdircache *));
1268 static void nfs_putdircache_unlocked __P((struct nfsnode *,
1269     struct nfsdircache *));
1270 
1271 static void
1272 nfs_unlinkdircache(np, ndp)
1273 	struct nfsnode *np;
1274 	struct nfsdircache *ndp;
1275 {
1276 
1277 	NFSDC_ASSERT_LOCKED(np);
1278 	KASSERT(ndp != &dzero);
1279 
1280 	if (LIST_NEXT(ndp, dc_hash) == (void *)-1)
1281 		return;
1282 
1283 	TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
1284 	LIST_REMOVE(ndp, dc_hash);
1285 	LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */
1286 
1287 	nfs_putdircache_unlocked(np, ndp);
1288 }
1289 
1290 void
1291 nfs_putdircache(np, ndp)
1292 	struct nfsnode *np;
1293 	struct nfsdircache *ndp;
1294 {
1295 	int ref;
1296 
1297 	if (ndp == &dzero)
1298 		return;
1299 
1300 	KASSERT(ndp->dc_refcnt > 0);
1301 	NFSDC_LOCK(np);
1302 	ref = --ndp->dc_refcnt;
1303 	NFSDC_UNLOCK(np);
1304 
1305 	if (ref == 0)
1306 		free(ndp, M_NFSDIROFF);
1307 }
1308 
1309 static void
1310 nfs_putdircache_unlocked(np, ndp)
1311 	struct nfsnode *np;
1312 	struct nfsdircache *ndp;
1313 {
1314 	int ref;
1315 
1316 	NFSDC_ASSERT_LOCKED(np);
1317 
1318 	if (ndp == &dzero)
1319 		return;
1320 
1321 	KASSERT(ndp->dc_refcnt > 0);
1322 	ref = --ndp->dc_refcnt;
1323 	if (ref == 0)
1324 		free(ndp, M_NFSDIROFF);
1325 }
1326 
1327 struct nfsdircache *
1328 nfs_searchdircache(vp, off, do32, hashent)
1329 	struct vnode *vp;
1330 	off_t off;
1331 	int do32;
1332 	int *hashent;
1333 {
1334 	struct nfsdirhashhead *ndhp;
1335 	struct nfsdircache *ndp = NULL;
1336 	struct nfsnode *np = VTONFS(vp);
1337 	unsigned ent;
1338 
1339 	/*
1340 	 * Zero is always a valid cookie.
1341 	 */
1342 	if (off == 0)
1343 		/* LINTED const cast away */
1344 		return (struct nfsdircache *)&dzero;
1345 
1346 	if (!np->n_dircache)
1347 		return NULL;
1348 
1349 	/*
1350 	 * We use a 32bit cookie as search key, directly reconstruct
1351 	 * the hashentry. Else use the hashfunction.
1352 	 */
1353 	if (do32) {
1354 		ent = (u_int32_t)off >> 24;
1355 		if (ent >= NFS_DIRHASHSIZ)
1356 			return NULL;
1357 		ndhp = &np->n_dircache[ent];
1358 	} else {
1359 		ndhp = NFSDIRHASH(np, off);
1360 	}
1361 
1362 	if (hashent)
1363 		*hashent = (int)(ndhp - np->n_dircache);
1364 
1365 	NFSDC_LOCK(np);
1366 	if (do32) {
1367 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1368 			if (ndp->dc_cookie32 == (u_int32_t)off) {
1369 				/*
1370 				 * An invalidated entry will become the
1371 				 * start of a new block fetched from
1372 				 * the server.
1373 				 */
1374 				if (ndp->dc_flags & NFSDC_INVALID) {
1375 					ndp->dc_blkcookie = ndp->dc_cookie;
1376 					ndp->dc_entry = 0;
1377 					ndp->dc_flags &= ~NFSDC_INVALID;
1378 				}
1379 				break;
1380 			}
1381 		}
1382 	} else {
1383 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1384 			if (ndp->dc_cookie == off)
1385 				break;
1386 		}
1387 	}
1388 	if (ndp != NULL)
1389 		ndp->dc_refcnt++;
1390 	NFSDC_UNLOCK(np);
1391 	return ndp;
1392 }
1393 
1394 
1395 struct nfsdircache *
1396 nfs_enterdircache(vp, off, blkoff, en, blkno)
1397 	struct vnode *vp;
1398 	off_t off, blkoff;
1399 	int en;
1400 	daddr_t blkno;
1401 {
1402 	struct nfsnode *np = VTONFS(vp);
1403 	struct nfsdirhashhead *ndhp;
1404 	struct nfsdircache *ndp = NULL;
1405 	struct nfsdircache *newndp = NULL;
1406 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1407 	int hashent, gen, overwrite;
1408 
1409 	/*
1410 	 * XXX refuse entries for offset 0. amd(8) erroneously sets
1411 	 * cookie 0 for the '.' entry, making this necessary. This
1412 	 * isn't so bad, as 0 is a special case anyway.
1413 	 */
1414 	if (off == 0)
1415 		/* LINTED const cast away */
1416 		return (struct nfsdircache *)&dzero;
1417 
1418 	if (!np->n_dircache)
1419 		/*
1420 		 * XXX would like to do this in nfs_nget but vtype
1421 		 * isn't known at that time.
1422 		 */
1423 		nfs_initdircache(vp);
1424 
1425 	if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens)
1426 		nfs_initdirxlatecookie(vp);
1427 
1428 retry:
1429 	ndp = nfs_searchdircache(vp, off, 0, &hashent);
1430 
1431 	NFSDC_LOCK(np);
1432 	if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) {
1433 		/*
1434 		 * Overwriting an old entry. Check if it's the same.
1435 		 * If so, just return. If not, remove the old entry.
1436 		 */
1437 		if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en)
1438 			goto done;
1439 		nfs_unlinkdircache(np, ndp);
1440 		nfs_putdircache_unlocked(np, ndp);
1441 		ndp = NULL;
1442 	}
1443 
1444 	ndhp = &np->n_dircache[hashent];
1445 
1446 	if (!ndp) {
1447 		if (newndp == NULL) {
1448 			NFSDC_UNLOCK(np);
1449 			newndp = malloc(sizeof(*ndp), M_NFSDIROFF, M_WAITOK);
1450 			newndp->dc_refcnt = 1;
1451 			LIST_NEXT(newndp, dc_hash) = (void *)-1;
1452 			goto retry;
1453 		}
1454 		ndp = newndp;
1455 		newndp = NULL;
1456 		overwrite = 0;
1457 		if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
1458 			/*
1459 			 * We're allocating a new entry, so bump the
1460 			 * generation number.
1461 			 */
1462 			gen = ++np->n_dirgens[hashent];
1463 			if (gen == 0) {
1464 				np->n_dirgens[hashent]++;
1465 				gen++;
1466 			}
1467 			ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff);
1468 		}
1469 	} else
1470 		overwrite = 1;
1471 
1472 	ndp->dc_cookie = off;
1473 	ndp->dc_blkcookie = blkoff;
1474 	ndp->dc_entry = en;
1475 	ndp->dc_flags = 0;
1476 
1477 	if (overwrite)
1478 		goto done;
1479 
1480 	/*
1481 	 * If the maximum directory cookie cache size has been reached
1482 	 * for this node, take one off the front. The idea is that
1483 	 * directories are typically read front-to-back once, so that
1484 	 * the oldest entries can be thrown away without much performance
1485 	 * loss.
1486 	 */
1487 	if (np->n_dircachesize == NFS_MAXDIRCACHE) {
1488 		nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain));
1489 	} else
1490 		np->n_dircachesize++;
1491 
1492 	KASSERT(ndp->dc_refcnt == 1);
1493 	LIST_INSERT_HEAD(ndhp, ndp, dc_hash);
1494 	TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain);
1495 	ndp->dc_refcnt++;
1496 done:
1497 	KASSERT(ndp->dc_refcnt > 0);
1498 	NFSDC_UNLOCK(np);
1499 	if (newndp)
1500 		nfs_putdircache(np, newndp);
1501 	return ndp;
1502 }
1503 
1504 void
1505 nfs_invaldircache(vp, flags)
1506 	struct vnode *vp;
1507 	int flags;
1508 {
1509 	struct nfsnode *np = VTONFS(vp);
1510 	struct nfsdircache *ndp = NULL;
1511 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1512 	const boolean_t forcefree = flags & NFS_INVALDIRCACHE_FORCE;
1513 
1514 #ifdef DIAGNOSTIC
1515 	if (vp->v_type != VDIR)
1516 		panic("nfs: invaldircache: not dir");
1517 #endif
1518 
1519 	if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0)
1520 		np->n_flag &= ~NEOFVALID;
1521 
1522 	if (!np->n_dircache)
1523 		return;
1524 
1525 	NFSDC_LOCK(np);
1526 	if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) {
1527 		while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) {
1528 			KASSERT(!forcefree || ndp->dc_refcnt == 1);
1529 			nfs_unlinkdircache(np, ndp);
1530 		}
1531 		np->n_dircachesize = 0;
1532 		if (forcefree && np->n_dirgens) {
1533 			FREE(np->n_dirgens, M_NFSDIROFF);
1534 			np->n_dirgens = NULL;
1535 		}
1536 	} else {
1537 		TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain)
1538 			ndp->dc_flags |= NFSDC_INVALID;
1539 	}
1540 
1541 	NFSDC_UNLOCK(np);
1542 }
1543 
1544 /*
1545  * Called once before VFS init to initialize shared and
1546  * server-specific data structures.
1547  */
1548 void
1549 nfs_init()
1550 {
1551 	nfsrtt.pos = 0;
1552 	rpc_vers = txdr_unsigned(RPC_VER2);
1553 	rpc_call = txdr_unsigned(RPC_CALL);
1554 	rpc_reply = txdr_unsigned(RPC_REPLY);
1555 	rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
1556 	rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
1557 	rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
1558 	rpc_autherr = txdr_unsigned(RPC_AUTHERR);
1559 	rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
1560 	rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
1561 	nfs_prog = txdr_unsigned(NFS_PROG);
1562 	nqnfs_prog = txdr_unsigned(NQNFS_PROG);
1563 	nfs_true = txdr_unsigned(TRUE);
1564 	nfs_false = txdr_unsigned(FALSE);
1565 	nfs_xdrneg1 = txdr_unsigned(-1);
1566 	nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
1567 	if (nfs_ticks < 1)
1568 		nfs_ticks = 1;
1569 #ifdef NFSSERVER
1570 	nfsrv_init(0);			/* Init server data structures */
1571 	nfsrv_initcache();		/* Init the server request cache */
1572 	pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript),
1573 	    0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr);
1574 #endif /* NFSSERVER */
1575 
1576 #if defined(NFSSERVER) || !defined(NFS_V2_ONLY)
1577 	/*
1578 	 * Initialize the nqnfs data structures.
1579 	 */
1580 	if (nqnfsstarttime == 0) {
1581 		nqnfsstarttime = boottime.tv_sec + nqsrv_maxlease
1582 			+ nqsrv_clockskew + nqsrv_writeslack;
1583 		NQLOADNOVRAM(nqnfsstarttime);
1584 		CIRCLEQ_INIT(&nqtimerhead);
1585 		nqfhhashtbl = hashinit(NQLCHSZ, HASH_LIST, M_NQLEASE,
1586 		    M_WAITOK, &nqfhhash);
1587 	}
1588 #endif
1589 
1590 	exithook_establish(nfs_exit, NULL);
1591 
1592 	/*
1593 	 * Initialize reply list and start timer
1594 	 */
1595 	TAILQ_INIT(&nfs_reqq);
1596 	nfs_timer(NULL);
1597 	MOWNER_ATTACH(&nfs_mowner);
1598 
1599 #ifdef NFS
1600 	/* Initialize the kqueue structures */
1601 	nfs_kqinit();
1602 	/* Initialize the iod structures */
1603 	nfs_iodinit();
1604 #endif
1605 }
1606 
1607 #ifdef NFS
1608 /*
1609  * Called once at VFS init to initialize client-specific data structures.
1610  */
1611 void
1612 nfs_vfs_init()
1613 {
1614 	nfs_nhinit();			/* Init the nfsnode table */
1615 	nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4);
1616 }
1617 
1618 void
1619 nfs_vfs_reinit()
1620 {
1621 	nfs_nhreinit();
1622 }
1623 
1624 void
1625 nfs_vfs_done()
1626 {
1627 	nfs_nhdone();
1628 }
1629 
1630 /*
1631  * Attribute cache routines.
1632  * nfs_loadattrcache() - loads or updates the cache contents from attributes
1633  *	that are on the mbuf list
1634  * nfs_getattrcache() - returns valid attributes if found in cache, returns
1635  *	error otherwise
1636  */
1637 
1638 /*
1639  * Load the attribute cache (that lives in the nfsnode entry) with
1640  * the values on the mbuf list and
1641  * Iff vap not NULL
1642  *    copy the attributes to *vaper
1643  */
1644 int
1645 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags)
1646 	struct vnode **vpp;
1647 	struct mbuf **mdp;
1648 	caddr_t *dposp;
1649 	struct vattr *vaper;
1650 	int flags;
1651 {
1652 	int32_t t1;
1653 	caddr_t cp2;
1654 	int error = 0;
1655 	struct mbuf *md;
1656 	int v3 = NFS_ISV3(*vpp);
1657 
1658 	md = *mdp;
1659 	t1 = (mtod(md, caddr_t) + md->m_len) - *dposp;
1660 	error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2);
1661 	if (error)
1662 		return (error);
1663 	return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags);
1664 }
1665 
1666 int
1667 nfs_loadattrcache(vpp, fp, vaper, flags)
1668 	struct vnode **vpp;
1669 	struct nfs_fattr *fp;
1670 	struct vattr *vaper;
1671 	int flags;
1672 {
1673 	struct vnode *vp = *vpp;
1674 	struct vattr *vap;
1675 	int v3 = NFS_ISV3(vp);
1676 	enum vtype vtyp;
1677 	u_short vmode;
1678 	struct timespec mtime;
1679 	struct timespec ctime;
1680 	struct vnode *nvp;
1681 	int32_t rdev;
1682 	struct nfsnode *np;
1683 	extern int (**spec_nfsv2nodeop_p) __P((void *));
1684 	uid_t uid;
1685 	gid_t gid;
1686 
1687 	if (v3) {
1688 		vtyp = nfsv3tov_type(fp->fa_type);
1689 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1690 		rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1),
1691 			fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2));
1692 		fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
1693 		fxdr_nfsv3time(&fp->fa3_ctime, &ctime);
1694 	} else {
1695 		vtyp = nfsv2tov_type(fp->fa_type);
1696 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1697 		if (vtyp == VNON || vtyp == VREG)
1698 			vtyp = IFTOVT(vmode);
1699 		rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
1700 		fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
1701 		ctime.tv_sec = fxdr_unsigned(u_int32_t,
1702 		    fp->fa2_ctime.nfsv2_sec);
1703 		ctime.tv_nsec = 0;
1704 
1705 		/*
1706 		 * Really ugly NFSv2 kludge.
1707 		 */
1708 		if (vtyp == VCHR && rdev == 0xffffffff)
1709 			vtyp = VFIFO;
1710 	}
1711 
1712 	vmode &= ALLPERMS;
1713 
1714 	/*
1715 	 * If v_type == VNON it is a new node, so fill in the v_type,
1716 	 * n_mtime fields. Check to see if it represents a special
1717 	 * device, and if so, check for a possible alias. Once the
1718 	 * correct vnode has been obtained, fill in the rest of the
1719 	 * information.
1720 	 */
1721 	np = VTONFS(vp);
1722 	if (vp->v_type == VNON) {
1723 		vp->v_type = vtyp;
1724 		if (vp->v_type == VFIFO) {
1725 			extern int (**fifo_nfsv2nodeop_p) __P((void *));
1726 			vp->v_op = fifo_nfsv2nodeop_p;
1727 		} else if (vp->v_type == VREG) {
1728 			lockinit(&np->n_commitlock, PINOD, "nfsclock", 0, 0);
1729 		} else if (vp->v_type == VCHR || vp->v_type == VBLK) {
1730 			vp->v_op = spec_nfsv2nodeop_p;
1731 			nvp = checkalias(vp, (dev_t)rdev, vp->v_mount);
1732 			if (nvp) {
1733 				/*
1734 				 * Discard unneeded vnode, but save its nfsnode.
1735 				 * Since the nfsnode does not have a lock, its
1736 				 * vnode lock has to be carried over.
1737 				 */
1738 				/*
1739 				 * XXX is the old node sure to be locked here?
1740 				 */
1741 				KASSERT(lockstatus(&vp->v_lock) ==
1742 				    LK_EXCLUSIVE);
1743 				nvp->v_data = vp->v_data;
1744 				vp->v_data = NULL;
1745 				VOP_UNLOCK(vp, 0);
1746 				vp->v_op = spec_vnodeop_p;
1747 				vrele(vp);
1748 				vgone(vp);
1749 				lockmgr(&nvp->v_lock, LK_EXCLUSIVE,
1750 				    &nvp->v_interlock);
1751 				/*
1752 				 * Reinitialize aliased node.
1753 				 */
1754 				np->n_vnode = nvp;
1755 				*vpp = vp = nvp;
1756 			}
1757 		}
1758 		np->n_mtime = mtime;
1759 	}
1760 	uid = fxdr_unsigned(uid_t, fp->fa_uid);
1761 	gid = fxdr_unsigned(gid_t, fp->fa_gid);
1762 	vap = np->n_vattr;
1763 
1764 	/*
1765 	 * Invalidate access cache if uid, gid, mode or ctime changed.
1766 	 */
1767 	if (np->n_accstamp != -1 &&
1768 	    (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode
1769 	    || timespeccmp(&ctime, &vap->va_ctime, !=)))
1770 		np->n_accstamp = -1;
1771 
1772 	vap->va_type = vtyp;
1773 	vap->va_mode = vmode;
1774 	vap->va_rdev = (dev_t)rdev;
1775 	vap->va_mtime = mtime;
1776 	vap->va_ctime = ctime;
1777 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
1778 	switch (vtyp) {
1779 	case VDIR:
1780 		vap->va_blocksize = NFS_DIRFRAGSIZ;
1781 		break;
1782 	case VBLK:
1783 		vap->va_blocksize = BLKDEV_IOSIZE;
1784 		break;
1785 	case VCHR:
1786 		vap->va_blocksize = MAXBSIZE;
1787 		break;
1788 	default:
1789 		vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize :
1790 		    fxdr_unsigned(int32_t, fp->fa2_blocksize);
1791 		break;
1792 	}
1793 	if (v3) {
1794 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1795 		vap->va_uid = uid;
1796 		vap->va_gid = gid;
1797 		vap->va_size = fxdr_hyper(&fp->fa3_size);
1798 		vap->va_bytes = fxdr_hyper(&fp->fa3_used);
1799 		vap->va_fileid = fxdr_unsigned(int32_t,
1800 		    fp->fa3_fileid.nfsuquad[1]);
1801 		fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
1802 		vap->va_flags = 0;
1803 		vap->va_filerev = 0;
1804 	} else {
1805 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1806 		vap->va_uid = uid;
1807 		vap->va_gid = gid;
1808 		vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
1809 		vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks)
1810 		    * NFS_FABLKSIZE;
1811 		vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
1812 		fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
1813 		vap->va_flags = 0;
1814 		vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
1815 		vap->va_filerev = 0;
1816 	}
1817 	if (vap->va_size != np->n_size) {
1818 		if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) {
1819 			vap->va_size = np->n_size;
1820 		} else {
1821 			np->n_size = vap->va_size;
1822 			if (vap->va_type == VREG) {
1823 				/*
1824 				 * we can't free pages if NAC_NOTRUNC because
1825 				 * the pages can be owned by ourselves.
1826 				 */
1827 				if (flags & NAC_NOTRUNC) {
1828 					np->n_flag |= NTRUNCDELAYED;
1829 				} else {
1830 					simple_lock(&vp->v_interlock);
1831 					(void)VOP_PUTPAGES(vp, 0,
1832 					    0, PGO_SYNCIO | PGO_CLEANIT |
1833 					    PGO_FREE | PGO_ALLPAGES);
1834 					uvm_vnp_setsize(vp, np->n_size);
1835 				}
1836 			}
1837 		}
1838 	}
1839 	np->n_attrstamp = mono_time.tv_sec;
1840 	if (vaper != NULL) {
1841 		memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(*vap));
1842 		if (np->n_flag & NCHG) {
1843 			if (np->n_flag & NACC)
1844 				vaper->va_atime = np->n_atim;
1845 			if (np->n_flag & NUPD)
1846 				vaper->va_mtime = np->n_mtim;
1847 		}
1848 	}
1849 	return (0);
1850 }
1851 
1852 /*
1853  * Check the time stamp
1854  * If the cache is valid, copy contents to *vap and return 0
1855  * otherwise return an error
1856  */
1857 int
1858 nfs_getattrcache(vp, vaper)
1859 	struct vnode *vp;
1860 	struct vattr *vaper;
1861 {
1862 	struct nfsnode *np = VTONFS(vp);
1863 	struct vattr *vap;
1864 
1865 	if (np->n_attrstamp == 0 ||
1866 	    (mono_time.tv_sec - np->n_attrstamp) >= NFS_ATTRTIMEO(np)) {
1867 		nfsstats.attrcache_misses++;
1868 		return (ENOENT);
1869 	}
1870 	nfsstats.attrcache_hits++;
1871 	vap = np->n_vattr;
1872 	if (vap->va_size != np->n_size) {
1873 		if (vap->va_type == VREG) {
1874 			if (np->n_flag & NMODIFIED) {
1875 				if (vap->va_size < np->n_size)
1876 					vap->va_size = np->n_size;
1877 				else
1878 					np->n_size = vap->va_size;
1879 			} else
1880 				np->n_size = vap->va_size;
1881 			uvm_vnp_setsize(vp, np->n_size);
1882 		} else
1883 			np->n_size = vap->va_size;
1884 	}
1885 	memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(struct vattr));
1886 	if (np->n_flag & NCHG) {
1887 		if (np->n_flag & NACC)
1888 			vaper->va_atime = np->n_atim;
1889 		if (np->n_flag & NUPD)
1890 			vaper->va_mtime = np->n_mtim;
1891 	}
1892 	return (0);
1893 }
1894 
1895 void
1896 nfs_delayedtruncate(vp)
1897 	struct vnode *vp;
1898 {
1899 	struct nfsnode *np = VTONFS(vp);
1900 
1901 	if (np->n_flag & NTRUNCDELAYED) {
1902 		np->n_flag &= ~NTRUNCDELAYED;
1903 		simple_lock(&vp->v_interlock);
1904 		(void)VOP_PUTPAGES(vp, 0,
1905 		    0, PGO_SYNCIO | PGO_CLEANIT | PGO_FREE | PGO_ALLPAGES);
1906 		uvm_vnp_setsize(vp, np->n_size);
1907 	}
1908 }
1909 
1910 #define	NFS_WCCKLUDGE_TIMEOUT	(24 * 60 * 60)	/* 1 day */
1911 #define	NFS_WCCKLUDGE(nmp, now) \
1912 	(((nmp)->nm_iflag & NFSMNT_WCCKLUDGE) && \
1913 	((now) - (nmp)->nm_wcckludgetime - NFS_WCCKLUDGE_TIMEOUT) < 0)
1914 
1915 /*
1916  * nfs_check_wccdata: check inaccurate wcc_data
1917  *
1918  * => return non-zero if we shouldn't trust the wcc_data.
1919  * => NFS_WCCKLUDGE_TIMEOUT is for the case that the server is "fixed".
1920  */
1921 
1922 int
1923 nfs_check_wccdata(struct nfsnode *np, const struct timespec *ctime,
1924     struct timespec *mtime, boolean_t docheck)
1925 {
1926 	int error = 0;
1927 
1928 #if !defined(NFS_V2_ONLY)
1929 
1930 	if (docheck) {
1931 		struct vnode *vp = NFSTOV(np);
1932 		struct nfsmount *nmp;
1933 		long now = mono_time.tv_sec;
1934 #if defined(DEBUG)
1935 		const char *reason = NULL; /* XXX: gcc */
1936 #endif
1937 
1938 		if (timespeccmp(&np->n_vattr->va_mtime, mtime, <=)) {
1939 #if defined(DEBUG)
1940 			reason = "mtime";
1941 #endif
1942 			error = EINVAL;
1943 		}
1944 
1945 		if (vp->v_type == VDIR &&
1946 		    timespeccmp(&np->n_vattr->va_ctime, ctime, <=)) {
1947 #if defined(DEBUG)
1948 			reason = "ctime";
1949 #endif
1950 			error = EINVAL;
1951 		}
1952 
1953 		nmp = VFSTONFS(vp->v_mount);
1954 		if (error) {
1955 
1956 			/*
1957 			 * despite of the fact that we've updated the file,
1958 			 * timestamps of the file were not updated as we
1959 			 * expected.
1960 			 * it means that the server has incompatible
1961 			 * semantics of timestamps or (more likely)
1962 			 * the server time is not precise enough to
1963 			 * track each modifications.
1964 			 * in that case, we disable wcc processing.
1965 			 *
1966 			 * yes, strictly speaking, we should disable all
1967 			 * caching.  it's a compromise.
1968 			 */
1969 
1970 			simple_lock(&nmp->nm_slock);
1971 #if defined(DEBUG)
1972 			if (!NFS_WCCKLUDGE(nmp, now)) {
1973 				printf("%s: inaccurate wcc data (%s) detected,"
1974 				    " disabling wcc\n",
1975 				    vp->v_mount->mnt_stat.f_mntfromname,
1976 				    reason);
1977 			}
1978 #endif
1979 			nmp->nm_iflag |= NFSMNT_WCCKLUDGE;
1980 			nmp->nm_wcckludgetime = now;
1981 			simple_unlock(&nmp->nm_slock);
1982 		} else if (NFS_WCCKLUDGE(nmp, now)) {
1983 			error = EPERM; /* XXX */
1984 		} else if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) {
1985 			simple_lock(&nmp->nm_slock);
1986 			if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) {
1987 #if defined(DEBUG)
1988 				printf("%s: re-enabling wcc\n",
1989 				    vp->v_mount->mnt_stat.f_mntfromname);
1990 #endif
1991 				nmp->nm_iflag &= ~NFSMNT_WCCKLUDGE;
1992 			}
1993 			simple_unlock(&nmp->nm_slock);
1994 		}
1995 	}
1996 
1997 #endif /* !defined(NFS_V2_ONLY) */
1998 
1999 	return error;
2000 }
2001 
2002 /*
2003  * Heuristic to see if the server XDR encodes directory cookies or not.
2004  * it is not supposed to, but a lot of servers may do this. Also, since
2005  * most/all servers will implement V2 as well, it is expected that they
2006  * may return just 32 bits worth of cookie information, so we need to
2007  * find out in which 32 bits this information is available. We do this
2008  * to avoid trouble with emulated binaries that can't handle 64 bit
2009  * directory offsets.
2010  */
2011 
2012 void
2013 nfs_cookieheuristic(vp, flagp, p, cred)
2014 	struct vnode *vp;
2015 	int *flagp;
2016 	struct proc *p;
2017 	struct ucred *cred;
2018 {
2019 	struct uio auio;
2020 	struct iovec aiov;
2021 	caddr_t buf, cp;
2022 	struct dirent *dp;
2023 	off_t *cookies = NULL, *cop;
2024 	int error, eof, nc, len;
2025 
2026 	MALLOC(buf, caddr_t, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK);
2027 
2028 	aiov.iov_base = buf;
2029 	aiov.iov_len = NFS_DIRFRAGSIZ;
2030 	auio.uio_iov = &aiov;
2031 	auio.uio_iovcnt = 1;
2032 	auio.uio_rw = UIO_READ;
2033 	auio.uio_segflg = UIO_SYSSPACE;
2034 	auio.uio_procp = NULL;
2035 	auio.uio_resid = NFS_DIRFRAGSIZ;
2036 	auio.uio_offset = 0;
2037 
2038 	error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc);
2039 
2040 	len = NFS_DIRFRAGSIZ - auio.uio_resid;
2041 	if (error || len == 0) {
2042 		FREE(buf, M_TEMP);
2043 		if (cookies)
2044 			free(cookies, M_TEMP);
2045 		return;
2046 	}
2047 
2048 	/*
2049 	 * Find the first valid entry and look at its offset cookie.
2050 	 */
2051 
2052 	cp = buf;
2053 	for (cop = cookies; len > 0; len -= dp->d_reclen) {
2054 		dp = (struct dirent *)cp;
2055 		if (dp->d_fileno != 0 && len >= dp->d_reclen) {
2056 			if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) {
2057 				*flagp |= NFSMNT_SWAPCOOKIE;
2058 				nfs_invaldircache(vp, 0);
2059 				nfs_vinvalbuf(vp, 0, cred, p, 1);
2060 			}
2061 			break;
2062 		}
2063 		cop++;
2064 		cp += dp->d_reclen;
2065 	}
2066 
2067 	FREE(buf, M_TEMP);
2068 	free(cookies, M_TEMP);
2069 }
2070 #endif /* NFS */
2071 
2072 /*
2073  * Set up nameidata for a lookup() call and do it.
2074  *
2075  * If pubflag is set, this call is done for a lookup operation on the
2076  * public filehandle. In that case we allow crossing mountpoints and
2077  * absolute pathnames. However, the caller is expected to check that
2078  * the lookup result is within the public fs, and deny access if
2079  * it is not.
2080  */
2081 int
2082 nfs_namei(ndp, fhp, len, slp, nam, mdp, dposp, retdirp, p, kerbflag, pubflag)
2083 	struct nameidata *ndp;
2084 	fhandle_t *fhp;
2085 	uint32_t len;
2086 	struct nfssvc_sock *slp;
2087 	struct mbuf *nam;
2088 	struct mbuf **mdp;
2089 	caddr_t *dposp;
2090 	struct vnode **retdirp;
2091 	struct proc *p;
2092 	int kerbflag, pubflag;
2093 {
2094 	int i, rem;
2095 	struct mbuf *md;
2096 	char *fromcp, *tocp, *cp;
2097 	struct iovec aiov;
2098 	struct uio auio;
2099 	struct vnode *dp;
2100 	int error, rdonly, linklen;
2101 	struct componentname *cnp = &ndp->ni_cnd;
2102 
2103 	*retdirp = (struct vnode *)0;
2104 
2105 	if ((len + 1) > MAXPATHLEN)
2106 		return (ENAMETOOLONG);
2107 	if (len == 0)
2108 		return (EACCES);
2109 	cnp->cn_pnbuf = PNBUF_GET();
2110 
2111 	/*
2112 	 * Copy the name from the mbuf list to ndp->ni_pnbuf
2113 	 * and set the various ndp fields appropriately.
2114 	 */
2115 	fromcp = *dposp;
2116 	tocp = cnp->cn_pnbuf;
2117 	md = *mdp;
2118 	rem = mtod(md, caddr_t) + md->m_len - fromcp;
2119 	for (i = 0; i < len; i++) {
2120 		while (rem == 0) {
2121 			md = md->m_next;
2122 			if (md == NULL) {
2123 				error = EBADRPC;
2124 				goto out;
2125 			}
2126 			fromcp = mtod(md, caddr_t);
2127 			rem = md->m_len;
2128 		}
2129 		if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
2130 			error = EACCES;
2131 			goto out;
2132 		}
2133 		*tocp++ = *fromcp++;
2134 		rem--;
2135 	}
2136 	*tocp = '\0';
2137 	*mdp = md;
2138 	*dposp = fromcp;
2139 	len = nfsm_rndup(len)-len;
2140 	if (len > 0) {
2141 		if (rem >= len)
2142 			*dposp += len;
2143 		else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
2144 			goto out;
2145 	}
2146 
2147 	/*
2148 	 * Extract and set starting directory.
2149 	 */
2150 	error = nfsrv_fhtovp(fhp, FALSE, &dp, ndp->ni_cnd.cn_cred, slp,
2151 	    nam, &rdonly, kerbflag, pubflag);
2152 	if (error)
2153 		goto out;
2154 	if (dp->v_type != VDIR) {
2155 		vrele(dp);
2156 		error = ENOTDIR;
2157 		goto out;
2158 	}
2159 
2160 	if (rdonly)
2161 		cnp->cn_flags |= RDONLY;
2162 
2163 	*retdirp = dp;
2164 
2165 	if (pubflag) {
2166 		/*
2167 		 * Oh joy. For WebNFS, handle those pesky '%' escapes,
2168 		 * and the 'native path' indicator.
2169 		 */
2170 		cp = PNBUF_GET();
2171 		fromcp = cnp->cn_pnbuf;
2172 		tocp = cp;
2173 		if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
2174 			switch ((unsigned char)*fromcp) {
2175 			case WEBNFS_NATIVE_CHAR:
2176 				/*
2177 				 * 'Native' path for us is the same
2178 				 * as a path according to the NFS spec,
2179 				 * just skip the escape char.
2180 				 */
2181 				fromcp++;
2182 				break;
2183 			/*
2184 			 * More may be added in the future, range 0x80-0xff
2185 			 */
2186 			default:
2187 				error = EIO;
2188 				PNBUF_PUT(cp);
2189 				goto out;
2190 			}
2191 		}
2192 		/*
2193 		 * Translate the '%' escapes, URL-style.
2194 		 */
2195 		while (*fromcp != '\0') {
2196 			if (*fromcp == WEBNFS_ESC_CHAR) {
2197 				if (fromcp[1] != '\0' && fromcp[2] != '\0') {
2198 					fromcp++;
2199 					*tocp++ = HEXSTRTOI(fromcp);
2200 					fromcp += 2;
2201 					continue;
2202 				} else {
2203 					error = ENOENT;
2204 					PNBUF_PUT(cp);
2205 					goto out;
2206 				}
2207 			} else
2208 				*tocp++ = *fromcp++;
2209 		}
2210 		*tocp = '\0';
2211 		PNBUF_PUT(cnp->cn_pnbuf);
2212 		cnp->cn_pnbuf = cp;
2213 	}
2214 
2215 	ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1;
2216 	ndp->ni_segflg = UIO_SYSSPACE;
2217 	ndp->ni_rootdir = rootvnode;
2218 
2219 	if (pubflag) {
2220 		ndp->ni_loopcnt = 0;
2221 		if (cnp->cn_pnbuf[0] == '/')
2222 			dp = rootvnode;
2223 	} else {
2224 		cnp->cn_flags |= NOCROSSMOUNT;
2225 	}
2226 
2227 	cnp->cn_proc = p;
2228 	VREF(dp);
2229 
2230     for (;;) {
2231 	cnp->cn_nameptr = cnp->cn_pnbuf;
2232 	ndp->ni_startdir = dp;
2233 	/*
2234 	 * And call lookup() to do the real work
2235 	 */
2236 	error = lookup(ndp);
2237 	if (error) {
2238 		PNBUF_PUT(cnp->cn_pnbuf);
2239 		return (error);
2240 	}
2241 	/*
2242 	 * Check for encountering a symbolic link
2243 	 */
2244 	if ((cnp->cn_flags & ISSYMLINK) == 0) {
2245 		if (cnp->cn_flags & (SAVENAME | SAVESTART))
2246 			cnp->cn_flags |= HASBUF;
2247 		else
2248 			PNBUF_PUT(cnp->cn_pnbuf);
2249 		return (0);
2250 	} else {
2251 		if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
2252 			VOP_UNLOCK(ndp->ni_dvp, 0);
2253 		if (!pubflag) {
2254 			error = EINVAL;
2255 			break;
2256 		}
2257 
2258 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
2259 			error = ELOOP;
2260 			break;
2261 		}
2262 		if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) {
2263 			error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred,
2264 			    cnp->cn_proc);
2265 			if (error != 0)
2266 				break;
2267 		}
2268 		if (ndp->ni_pathlen > 1)
2269 			cp = PNBUF_GET();
2270 		else
2271 			cp = cnp->cn_pnbuf;
2272 		aiov.iov_base = cp;
2273 		aiov.iov_len = MAXPATHLEN;
2274 		auio.uio_iov = &aiov;
2275 		auio.uio_iovcnt = 1;
2276 		auio.uio_offset = 0;
2277 		auio.uio_rw = UIO_READ;
2278 		auio.uio_segflg = UIO_SYSSPACE;
2279 		auio.uio_procp = NULL;
2280 		auio.uio_resid = MAXPATHLEN;
2281 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
2282 		if (error) {
2283 		badlink:
2284 			if (ndp->ni_pathlen > 1)
2285 				PNBUF_PUT(cp);
2286 			break;
2287 		}
2288 		linklen = MAXPATHLEN - auio.uio_resid;
2289 		if (linklen == 0) {
2290 			error = ENOENT;
2291 			goto badlink;
2292 		}
2293 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
2294 			error = ENAMETOOLONG;
2295 			goto badlink;
2296 		}
2297 		if (ndp->ni_pathlen > 1) {
2298 			memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
2299 			PNBUF_PUT(cnp->cn_pnbuf);
2300 			cnp->cn_pnbuf = cp;
2301 		} else
2302 			cnp->cn_pnbuf[linklen] = '\0';
2303 		ndp->ni_pathlen += linklen;
2304 		vput(ndp->ni_vp);
2305 		dp = ndp->ni_dvp;
2306 		/*
2307 		 * Check if root directory should replace current directory.
2308 		 */
2309 		if (cnp->cn_pnbuf[0] == '/') {
2310 			vrele(dp);
2311 			dp = ndp->ni_rootdir;
2312 			VREF(dp);
2313 		}
2314 	}
2315    }
2316 	vrele(ndp->ni_dvp);
2317 	vput(ndp->ni_vp);
2318 	ndp->ni_vp = NULL;
2319 out:
2320 	PNBUF_PUT(cnp->cn_pnbuf);
2321 	return (error);
2322 }
2323 
2324 /*
2325  * A fiddled version of m_adj() that ensures null fill to a 32-bit
2326  * boundary and only trims off the back end
2327  *
2328  * 1. trim off 'len' bytes as m_adj(mp, -len).
2329  * 2. add zero-padding 'nul' bytes at the end of the mbuf chain.
2330  */
2331 void
2332 nfs_zeropad(mp, len, nul)
2333 	struct mbuf *mp;
2334 	int len;
2335 	int nul;
2336 {
2337 	struct mbuf *m;
2338 	int count;
2339 
2340 	/*
2341 	 * Trim from tail.  Scan the mbuf chain,
2342 	 * calculating its length and finding the last mbuf.
2343 	 * If the adjustment only affects this mbuf, then just
2344 	 * adjust and return.  Otherwise, rescan and truncate
2345 	 * after the remaining size.
2346 	 */
2347 	count = 0;
2348 	m = mp;
2349 	for (;;) {
2350 		count += m->m_len;
2351 		if (m->m_next == NULL)
2352 			break;
2353 		m = m->m_next;
2354 	}
2355 
2356 	KDASSERT(count >= len);
2357 
2358 	if (m->m_len >= len) {
2359 		m->m_len -= len;
2360 	} else {
2361 		count -= len;
2362 		/*
2363 		 * Correct length for chain is "count".
2364 		 * Find the mbuf with last data, adjust its length,
2365 		 * and toss data from remaining mbufs on chain.
2366 		 */
2367 		for (m = mp; m; m = m->m_next) {
2368 			if (m->m_len >= count) {
2369 				m->m_len = count;
2370 				break;
2371 			}
2372 			count -= m->m_len;
2373 		}
2374 		m_freem(m->m_next);
2375 		m->m_next = NULL;
2376 	}
2377 
2378 	KDASSERT(m->m_next == NULL);
2379 
2380 	/*
2381 	 * zero-padding.
2382 	 */
2383 	if (nul > 0) {
2384 		char *cp;
2385 		int i;
2386 
2387 		if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) {
2388 			struct mbuf *n;
2389 
2390 			KDASSERT(MLEN >= nul);
2391 			n = m_get(M_WAIT, MT_DATA);
2392 			MCLAIM(n, &nfs_mowner);
2393 			n->m_len = nul;
2394 			n->m_next = NULL;
2395 			m->m_next = n;
2396 			cp = mtod(n, caddr_t);
2397 		} else {
2398 			cp = mtod(m, caddr_t) + m->m_len;
2399 			m->m_len += nul;
2400 		}
2401 		for (i = 0; i < nul; i++)
2402 			*cp++ = '\0';
2403 	}
2404 	return;
2405 }
2406 
2407 /*
2408  * Make these functions instead of macros, so that the kernel text size
2409  * doesn't get too big...
2410  */
2411 void
2412 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp)
2413 	struct nfsrv_descript *nfsd;
2414 	int before_ret;
2415 	struct vattr *before_vap;
2416 	int after_ret;
2417 	struct vattr *after_vap;
2418 	struct mbuf **mbp;
2419 	char **bposp;
2420 {
2421 	struct mbuf *mb = *mbp;
2422 	char *bpos = *bposp;
2423 	u_int32_t *tl;
2424 
2425 	if (before_ret) {
2426 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2427 		*tl = nfs_false;
2428 	} else {
2429 		nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED);
2430 		*tl++ = nfs_true;
2431 		txdr_hyper(before_vap->va_size, tl);
2432 		tl += 2;
2433 		txdr_nfsv3time(&(before_vap->va_mtime), tl);
2434 		tl += 2;
2435 		txdr_nfsv3time(&(before_vap->va_ctime), tl);
2436 	}
2437 	*bposp = bpos;
2438 	*mbp = mb;
2439 	nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp);
2440 }
2441 
2442 void
2443 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp)
2444 	struct nfsrv_descript *nfsd;
2445 	int after_ret;
2446 	struct vattr *after_vap;
2447 	struct mbuf **mbp;
2448 	char **bposp;
2449 {
2450 	struct mbuf *mb = *mbp;
2451 	char *bpos = *bposp;
2452 	u_int32_t *tl;
2453 	struct nfs_fattr *fp;
2454 
2455 	if (after_ret) {
2456 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2457 		*tl = nfs_false;
2458 	} else {
2459 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR);
2460 		*tl++ = nfs_true;
2461 		fp = (struct nfs_fattr *)tl;
2462 		nfsm_srvfattr(nfsd, after_vap, fp);
2463 	}
2464 	*mbp = mb;
2465 	*bposp = bpos;
2466 }
2467 
2468 void
2469 nfsm_srvfattr(nfsd, vap, fp)
2470 	struct nfsrv_descript *nfsd;
2471 	struct vattr *vap;
2472 	struct nfs_fattr *fp;
2473 {
2474 
2475 	fp->fa_nlink = txdr_unsigned(vap->va_nlink);
2476 	fp->fa_uid = txdr_unsigned(vap->va_uid);
2477 	fp->fa_gid = txdr_unsigned(vap->va_gid);
2478 	if (nfsd->nd_flag & ND_NFSV3) {
2479 		fp->fa_type = vtonfsv3_type(vap->va_type);
2480 		fp->fa_mode = vtonfsv3_mode(vap->va_mode);
2481 		txdr_hyper(vap->va_size, &fp->fa3_size);
2482 		txdr_hyper(vap->va_bytes, &fp->fa3_used);
2483 		fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev));
2484 		fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev));
2485 		fp->fa3_fsid.nfsuquad[0] = 0;
2486 		fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid);
2487 		fp->fa3_fileid.nfsuquad[0] = 0;
2488 		fp->fa3_fileid.nfsuquad[1] = txdr_unsigned(vap->va_fileid);
2489 		txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime);
2490 		txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime);
2491 		txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime);
2492 	} else {
2493 		fp->fa_type = vtonfsv2_type(vap->va_type);
2494 		fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
2495 		fp->fa2_size = txdr_unsigned(vap->va_size);
2496 		fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize);
2497 		if (vap->va_type == VFIFO)
2498 			fp->fa2_rdev = 0xffffffff;
2499 		else
2500 			fp->fa2_rdev = txdr_unsigned(vap->va_rdev);
2501 		fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE);
2502 		fp->fa2_fsid = txdr_unsigned(vap->va_fsid);
2503 		fp->fa2_fileid = txdr_unsigned(vap->va_fileid);
2504 		txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime);
2505 		txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime);
2506 		txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime);
2507 	}
2508 }
2509 
2510 /*
2511  * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked)
2512  * 	- look up fsid in mount list (if not found ret error)
2513  *	- get vp and export rights by calling VFS_FHTOVP()
2514  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2515  *	- if not lockflag unlock it with VOP_UNLOCK()
2516  */
2517 int
2518 nfsrv_fhtovp(fhp, lockflag, vpp, cred, slp, nam, rdonlyp, kerbflag, pubflag)
2519 	fhandle_t *fhp;
2520 	int lockflag;
2521 	struct vnode **vpp;
2522 	struct ucred *cred;
2523 	struct nfssvc_sock *slp;
2524 	struct mbuf *nam;
2525 	int *rdonlyp;
2526 	int kerbflag;
2527 {
2528 	struct mount *mp;
2529 	int i;
2530 	struct ucred *credanon;
2531 	int error, exflags;
2532 	struct sockaddr_in *saddr;
2533 
2534 	*vpp = (struct vnode *)0;
2535 
2536 	if (nfs_ispublicfh(fhp)) {
2537 		if (!pubflag || !nfs_pub.np_valid)
2538 			return (ESTALE);
2539 		fhp = &nfs_pub.np_handle;
2540 	}
2541 
2542 	mp = vfs_getvfs(&fhp->fh_fsid);
2543 	if (!mp)
2544 		return (ESTALE);
2545 	error = VFS_CHECKEXP(mp, nam, &exflags, &credanon);
2546 	if (error)
2547 		return (error);
2548 	error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp);
2549 	if (error)
2550 		return (error);
2551 
2552 	if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) {
2553 		saddr = mtod(nam, struct sockaddr_in *);
2554 		if ((saddr->sin_family == AF_INET) &&
2555 		    ntohs(saddr->sin_port) >= IPPORT_RESERVED) {
2556 			vput(*vpp);
2557 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2558 		}
2559 #ifdef INET6
2560 		if ((saddr->sin_family == AF_INET6) &&
2561 		    ntohs(saddr->sin_port) >= IPV6PORT_RESERVED) {
2562 			vput(*vpp);
2563 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2564 		}
2565 #endif
2566 	}
2567 	/*
2568 	 * Check/setup credentials.
2569 	 */
2570 	if (exflags & MNT_EXKERB) {
2571 		if (!kerbflag) {
2572 			vput(*vpp);
2573 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2574 		}
2575 	} else if (kerbflag) {
2576 		vput(*vpp);
2577 		return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2578 	} else if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
2579 		cred->cr_uid = credanon->cr_uid;
2580 		cred->cr_gid = credanon->cr_gid;
2581 		for (i = 0; i < credanon->cr_ngroups && i < NGROUPS; i++)
2582 			cred->cr_groups[i] = credanon->cr_groups[i];
2583 		cred->cr_ngroups = i;
2584 	}
2585 	if (exflags & MNT_EXRDONLY)
2586 		*rdonlyp = 1;
2587 	else
2588 		*rdonlyp = 0;
2589 	if (!lockflag)
2590 		VOP_UNLOCK(*vpp, 0);
2591 	return (0);
2592 }
2593 
2594 /*
2595  * WebNFS: check if a filehandle is a public filehandle. For v3, this
2596  * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has
2597  * transformed this to all zeroes in both cases, so check for it.
2598  */
2599 int
2600 nfs_ispublicfh(fhp)
2601 	fhandle_t *fhp;
2602 {
2603 	char *cp = (char *)fhp;
2604 	int i;
2605 
2606 	for (i = 0; i < NFSX_V3FH; i++)
2607 		if (*cp++ != 0)
2608 			return (FALSE);
2609 	return (TRUE);
2610 }
2611 
2612 /*
2613  * This function compares two net addresses by family and returns TRUE
2614  * if they are the same host.
2615  * If there is any doubt, return FALSE.
2616  * The AF_INET family is handled as a special case so that address mbufs
2617  * don't need to be saved to store "struct in_addr", which is only 4 bytes.
2618  */
2619 int
2620 netaddr_match(family, haddr, nam)
2621 	int family;
2622 	union nethostaddr *haddr;
2623 	struct mbuf *nam;
2624 {
2625 	struct sockaddr_in *inetaddr;
2626 
2627 	switch (family) {
2628 	case AF_INET:
2629 		inetaddr = mtod(nam, struct sockaddr_in *);
2630 		if (inetaddr->sin_family == AF_INET &&
2631 		    inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
2632 			return (1);
2633 		break;
2634 #ifdef INET6
2635 	case AF_INET6:
2636 	    {
2637 		struct sockaddr_in6 *sin6_1, *sin6_2;
2638 
2639 		sin6_1 = mtod(nam, struct sockaddr_in6 *);
2640 		sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *);
2641 		if (sin6_1->sin6_family == AF_INET6 &&
2642 		    IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr))
2643 			return 1;
2644 	    }
2645 #endif
2646 #ifdef ISO
2647 	case AF_ISO:
2648 	    {
2649 		struct sockaddr_iso *isoaddr1, *isoaddr2;
2650 
2651 		isoaddr1 = mtod(nam, struct sockaddr_iso *);
2652 		isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *);
2653 		if (isoaddr1->siso_family == AF_ISO &&
2654 		    isoaddr1->siso_nlen > 0 &&
2655 		    isoaddr1->siso_nlen == isoaddr2->siso_nlen &&
2656 		    SAME_ISOADDR(isoaddr1, isoaddr2))
2657 			return (1);
2658 		break;
2659 	    }
2660 #endif	/* ISO */
2661 	default:
2662 		break;
2663 	};
2664 	return (0);
2665 }
2666 
2667 /*
2668  * The write verifier has changed (probably due to a server reboot), so all
2669  * PG_NEEDCOMMIT pages will have to be written again. Since they are marked
2670  * as dirty or are being written out just now, all this takes is clearing
2671  * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for
2672  * the mount point.
2673  */
2674 void
2675 nfs_clearcommit(mp)
2676 	struct mount *mp;
2677 {
2678 	struct vnode *vp;
2679 	struct nfsnode *np;
2680 	struct vm_page *pg;
2681 	struct nfsmount *nmp = VFSTONFS(mp);
2682 
2683 	lockmgr(&nmp->nm_writeverflock, LK_EXCLUSIVE, NULL);
2684 
2685 	LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2686 		KASSERT(vp->v_mount == mp);
2687 		if (vp->v_type != VREG)
2688 			continue;
2689 		np = VTONFS(vp);
2690 		np->n_pushlo = np->n_pushhi = np->n_pushedlo =
2691 		    np->n_pushedhi = 0;
2692 		np->n_commitflags &=
2693 		    ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID);
2694 		simple_lock(&vp->v_uobj.vmobjlock);
2695 		TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq) {
2696 			pg->flags &= ~PG_NEEDCOMMIT;
2697 		}
2698 		simple_unlock(&vp->v_uobj.vmobjlock);
2699 	}
2700 	simple_lock(&nmp->nm_slock);
2701 	nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF;
2702 	simple_unlock(&nmp->nm_slock);
2703 	lockmgr(&nmp->nm_writeverflock, LK_RELEASE, NULL);
2704 }
2705 
2706 void
2707 nfs_merge_commit_ranges(vp)
2708 	struct vnode *vp;
2709 {
2710 	struct nfsnode *np = VTONFS(vp);
2711 
2712 	KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID);
2713 
2714 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2715 		np->n_pushedlo = np->n_pushlo;
2716 		np->n_pushedhi = np->n_pushhi;
2717 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2718 	} else {
2719 		if (np->n_pushlo < np->n_pushedlo)
2720 			np->n_pushedlo = np->n_pushlo;
2721 		if (np->n_pushhi > np->n_pushedhi)
2722 			np->n_pushedhi = np->n_pushhi;
2723 	}
2724 
2725 	np->n_pushlo = np->n_pushhi = 0;
2726 	np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID;
2727 
2728 #ifdef NFS_DEBUG_COMMIT
2729 	printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2730 	    (unsigned)np->n_pushedhi);
2731 #endif
2732 }
2733 
2734 int
2735 nfs_in_committed_range(vp, off, len)
2736 	struct vnode *vp;
2737 	off_t off, len;
2738 {
2739 	struct nfsnode *np = VTONFS(vp);
2740 	off_t lo, hi;
2741 
2742 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2743 		return 0;
2744 	lo = off;
2745 	hi = lo + len;
2746 
2747 	return (lo >= np->n_pushedlo && hi <= np->n_pushedhi);
2748 }
2749 
2750 int
2751 nfs_in_tobecommitted_range(vp, off, len)
2752 	struct vnode *vp;
2753 	off_t off, len;
2754 {
2755 	struct nfsnode *np = VTONFS(vp);
2756 	off_t lo, hi;
2757 
2758 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2759 		return 0;
2760 	lo = off;
2761 	hi = lo + len;
2762 
2763 	return (lo >= np->n_pushlo && hi <= np->n_pushhi);
2764 }
2765 
2766 void
2767 nfs_add_committed_range(vp, off, len)
2768 	struct vnode *vp;
2769 	off_t off, len;
2770 {
2771 	struct nfsnode *np = VTONFS(vp);
2772 	off_t lo, hi;
2773 
2774 	lo = off;
2775 	hi = lo + len;
2776 
2777 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2778 		np->n_pushedlo = lo;
2779 		np->n_pushedhi = hi;
2780 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2781 	} else {
2782 		if (hi > np->n_pushedhi)
2783 			np->n_pushedhi = hi;
2784 		if (lo < np->n_pushedlo)
2785 			np->n_pushedlo = lo;
2786 	}
2787 #ifdef NFS_DEBUG_COMMIT
2788 	printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2789 	    (unsigned)np->n_pushedhi);
2790 #endif
2791 }
2792 
2793 void
2794 nfs_del_committed_range(vp, off, len)
2795 	struct vnode *vp;
2796 	off_t off, len;
2797 {
2798 	struct nfsnode *np = VTONFS(vp);
2799 	off_t lo, hi;
2800 
2801 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2802 		return;
2803 
2804 	lo = off;
2805 	hi = lo + len;
2806 
2807 	if (lo > np->n_pushedhi || hi < np->n_pushedlo)
2808 		return;
2809 	if (lo <= np->n_pushedlo)
2810 		np->n_pushedlo = hi;
2811 	else if (hi >= np->n_pushedhi)
2812 		np->n_pushedhi = lo;
2813 	else {
2814 		/*
2815 		 * XXX There's only one range. If the deleted range
2816 		 * is in the middle, pick the largest of the
2817 		 * contiguous ranges that it leaves.
2818 		 */
2819 		if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi))
2820 			np->n_pushedhi = lo;
2821 		else
2822 			np->n_pushedlo = hi;
2823 	}
2824 #ifdef NFS_DEBUG_COMMIT
2825 	printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2826 	    (unsigned)np->n_pushedhi);
2827 #endif
2828 }
2829 
2830 void
2831 nfs_add_tobecommitted_range(vp, off, len)
2832 	struct vnode *vp;
2833 	off_t off, len;
2834 {
2835 	struct nfsnode *np = VTONFS(vp);
2836 	off_t lo, hi;
2837 
2838 	lo = off;
2839 	hi = lo + len;
2840 
2841 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) {
2842 		np->n_pushlo = lo;
2843 		np->n_pushhi = hi;
2844 		np->n_commitflags |= NFS_COMMIT_PUSH_VALID;
2845 	} else {
2846 		if (lo < np->n_pushlo)
2847 			np->n_pushlo = lo;
2848 		if (hi > np->n_pushhi)
2849 			np->n_pushhi = hi;
2850 	}
2851 #ifdef NFS_DEBUG_COMMIT
2852 	printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2853 	    (unsigned)np->n_pushhi);
2854 #endif
2855 }
2856 
2857 void
2858 nfs_del_tobecommitted_range(vp, off, len)
2859 	struct vnode *vp;
2860 	off_t off, len;
2861 {
2862 	struct nfsnode *np = VTONFS(vp);
2863 	off_t lo, hi;
2864 
2865 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2866 		return;
2867 
2868 	lo = off;
2869 	hi = lo + len;
2870 
2871 	if (lo > np->n_pushhi || hi < np->n_pushlo)
2872 		return;
2873 
2874 	if (lo <= np->n_pushlo)
2875 		np->n_pushlo = hi;
2876 	else if (hi >= np->n_pushhi)
2877 		np->n_pushhi = lo;
2878 	else {
2879 		/*
2880 		 * XXX There's only one range. If the deleted range
2881 		 * is in the middle, pick the largest of the
2882 		 * contiguous ranges that it leaves.
2883 		 */
2884 		if ((np->n_pushlo - lo) > (hi - np->n_pushhi))
2885 			np->n_pushhi = lo;
2886 		else
2887 			np->n_pushlo = hi;
2888 	}
2889 #ifdef NFS_DEBUG_COMMIT
2890 	printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2891 	    (unsigned)np->n_pushhi);
2892 #endif
2893 }
2894 
2895 /*
2896  * Map errnos to NFS error numbers. For Version 3 also filter out error
2897  * numbers not specified for the associated procedure.
2898  */
2899 int
2900 nfsrv_errmap(nd, err)
2901 	struct nfsrv_descript *nd;
2902 	int err;
2903 {
2904 	const short *defaulterrp, *errp;
2905 
2906 	if (nd->nd_flag & ND_NFSV3) {
2907 	    if (nd->nd_procnum <= NFSPROC_COMMIT) {
2908 		errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
2909 		while (*++errp) {
2910 			if (*errp == err)
2911 				return (err);
2912 			else if (*errp > err)
2913 				break;
2914 		}
2915 		return ((int)*defaulterrp);
2916 	    } else
2917 		return (err & 0xffff);
2918 	}
2919 	if (err <= ELAST)
2920 		return ((int)nfsrv_v2errmap[err - 1]);
2921 	return (NFSERR_IO);
2922 }
2923 
2924 /*
2925  * Sort the group list in increasing numerical order.
2926  * (Insertion sort by Chris Torek, who was grossed out by the bubble sort
2927  *  that used to be here.)
2928  */
2929 void
2930 nfsrvw_sort(list, num)
2931         gid_t *list;
2932         int num;
2933 {
2934 	int i, j;
2935 	gid_t v;
2936 
2937 	/* Insertion sort. */
2938 	for (i = 1; i < num; i++) {
2939 		v = list[i];
2940 		/* find correct slot for value v, moving others up */
2941 		for (j = i; --j >= 0 && v < list[j];)
2942 			list[j + 1] = list[j];
2943 		list[j + 1] = v;
2944 	}
2945 }
2946 
2947 /*
2948  * copy credentials making sure that the result can be compared with memcmp().
2949  */
2950 void
2951 nfsrv_setcred(incred, outcred)
2952 	struct ucred *incred, *outcred;
2953 {
2954 	int i;
2955 
2956 	memset((caddr_t)outcred, 0, sizeof (struct ucred));
2957 	outcred->cr_ref = 1;
2958 	outcred->cr_uid = incred->cr_uid;
2959 	outcred->cr_gid = incred->cr_gid;
2960 	outcred->cr_ngroups = incred->cr_ngroups;
2961 	for (i = 0; i < incred->cr_ngroups; i++)
2962 		outcred->cr_groups[i] = incred->cr_groups[i];
2963 	nfsrvw_sort(outcred->cr_groups, outcred->cr_ngroups);
2964 }
2965 
2966 u_int32_t
2967 nfs_getxid()
2968 {
2969 	static u_int32_t base;
2970 	static u_int32_t nfs_xid = 0;
2971 	static struct simplelock nfs_xidlock = SIMPLELOCK_INITIALIZER;
2972 	u_int32_t newxid;
2973 
2974 	simple_lock(&nfs_xidlock);
2975 	/*
2976 	 * derive initial xid from system time
2977 	 * XXX time is invalid if root not yet mounted
2978 	 */
2979 	if (__predict_false(!base && (rootvp))) {
2980 		struct timeval tv;
2981 
2982 		microtime(&tv);
2983 		base = tv.tv_sec << 12;
2984 		nfs_xid = base;
2985 	}
2986 
2987 	/*
2988 	 * Skip zero xid if it should ever happen.
2989 	 */
2990 	if (__predict_false(++nfs_xid == 0))
2991 		nfs_xid++;
2992 	newxid = nfs_xid;
2993 	simple_unlock(&nfs_xidlock);
2994 
2995 	return txdr_unsigned(newxid);
2996 }
2997 
2998 /*
2999  * assign a new xid for existing request.
3000  * used for NFSERR_JUKEBOX handling.
3001  */
3002 void
3003 nfs_renewxid(struct nfsreq *req)
3004 {
3005 	u_int32_t xid;
3006 	int off;
3007 
3008 	xid = nfs_getxid();
3009 	if (req->r_nmp->nm_sotype == SOCK_STREAM)
3010 		off = sizeof(u_int32_t); /* RPC record mark */
3011 	else
3012 		off = 0;
3013 
3014 	m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid);
3015 	req->r_xid = xid;
3016 }
3017